
This article is part of the guide The plan that holds: Microsoft Project best practices, symptom by symptom, nineteen things you can see in a plan and the move that reveals each one in thirty seconds.
Open any plan that has been through three replans and look at the links. Some of them describe the work: the walls cannot go up before the foundation has cured. Others describe a preference somebody had in January: the training was going to happen after the pilot because that seemed tidy. And a few describe nothing at all — they were dragged in the Gantt view to make a bar land on a date the client liked.
The three kinds look identical on screen. They behave very differently when the plan moves, and telling them apart is most of what there is to know about task dependencies. This article goes through the four link types, the number that rides on a link, the difference between a hard and a soft dependency, and the one place a link should never be drawn.
In short
- A dependency is a statement about the work: what one task needs from another before it can start or finish. If you cannot say what B needs from A, there is no dependency, only a wish about dates.
- There are four link types: finish-to-start, start-to-start, finish-to-finish, start-to-finish. Nine links out of ten are finish-to-start; start-to-finish is rare and usually a sign the tasks were written the wrong way round.
- Lead and lag are the number on the link:
FS+3dwaits three days after A finishes,FS-2dlets B start two days before. - Mandatory (hard) dependencies come from physics, safety or contract and cannot be changed. Discretionary (soft) ones are a choice, and should be labelled as such, because they are the first thing to remove when the schedule needs compressing.
- Links go on detail tasks, never on summary tasks. A summary is a container; linking containers hides the real logic and makes the critical path unreadable.
A dependency is a fact about the work, not about the calendar
The word makes it sound abstract, so start from the concrete version. A dependency between two tasks says that B needs something from A: a finished deliverable, a decision, a resource being released, a state of the site. That need is what the link models. The dates come afterwards, calculated by the scheduling engine from the links, the durations and the calendars.
This order matters because it is the one that survives change. When a task slips, a plan built on real dependencies moves the right things by the right amount and leaves everything else alone. A plan whose links were drawn to hit dates moves everything, or nothing, and the planner ends up fixing it by hand — which usually means typing dates, which usually means constraints, which is how a schedule stops being a schedule and becomes a list of promises.
The test is one question, asked of every link: what does B need from A? If the answer is a thing — the cured concrete, the signed drawing, the test report — the link is right. If the answer is “it seemed sensible” the link is a soft dependency and should be marked as one (below). If there is no answer, delete it.
The four types of task dependencies
Every scheduling tool, Microsoft Project included, offers the same four. In Microsoft’s own wording, with A the task that is depended on and B the dependent task:

Finish-to-start is the default, and for a good reason: most work really does wait for something to be finished. When in doubt, it is the right one.
Start-to-start models two things that run in parallel but cannot both start from zero: the second crew needs the first to have begun. It almost always carries a lag — hauling starts a day after digging — and it is the honest way to overlap tasks instead of splitting them into artificial halves.
Finish-to-finish is the mirror: B can run alongside A but cannot be declared done first. The user manual is the classic case — you can write while testing runs, but the last defects change the text, so the manual finishes after the tests do.
Start-to-finish is the one people learn for the exam and then never meet. Its legitimate use is a hand-over: the old system runs until the new one starts, so the start of A decides the finish of B. If you find an SF link in a plan and it is not a hand-over of that kind, the two tasks were probably written in the wrong order and an FS link between them the other way round says the same thing more clearly.
Lead and lag: the number on the link
A link on its own says “immediately”. Most real dependencies have a gap or an overlap, and that is what lag and lead are for.
- Lag is a delay after the link is satisfied: concrete needs to cure before the walls go up, so the link is finish-to-start plus three days. In Project you type it in the Lag column, or straight into the Predecessors column as
4FS+3d. - Lead is an overlap: the walls can start two days before the last pour is finished because the first section has cured. In Project lead is a negative lag — Microsoft’s page says it plainly: “type a negative value in the Lag column, such as –2” — so the link reads
4FS-2d.
The number is part of the dependency and deserves the same question as the link: where does the three days come from? Curing time is a fact. “Two days to be safe” is a buffer hiding inside a link, and buffers hidden inside links are invisible when you look for where the float went. If you want a buffer, make it a task with a name, so everybody can see it and somebody can decide to spend it.
Hard and soft: mandatory, discretionary, external
Two links can be drawn identically and mean different things. The distinction that matters when the schedule comes under pressure is where the dependency comes from.
- Mandatory (hard) dependencies are imposed by the nature of the work: physics, safety, a contract clause, a regulation. The electrical installation must be finished before the safety tests can begin; there is no version of the project in which that is not true. A hard link cannot be removed to gain time. It can only be shortened by changing the work itself.
- Discretionary (soft) dependencies are a choice of the team: “we prefer to run the training after the pilot”, “we would rather finish design before starting procurement”. They are sensible, they are often right, and they are the first thing to look at when the finish date does not fit, because removing one costs nothing but a preference.
- External dependencies come from outside the project: a permit from the council, a component from a supplier, a decision from another department. They behave like hard links you do not control, and they deserve a milestone of their own in the plan so the wait is visible and dated.
The practical problem is that Project draws all three the same way. The fix is cheap: a text or flag custom field on the task, or a note on the link, that says hard, soft or external. When a replan comes — and it comes — the soft links are the ones you filter for first, and you do not want to be reconstructing from memory which ones they were. A plan where every link is treated as hard cannot be compressed except by cutting scope; a plan where every link is treated as soft is not a plan.
Where the link goes: on the detail task, never on the summary
This is the mistake that costs the most and looks the most reasonable. Two phases, Design and Build; Build comes after Design; so somebody links the two summary tasks and moves on.

What goes wrong is not visible on the first day. A summary task is a container: it takes its start, finish, duration and progress from the tasks inside it. Put a link on the container and you have declared that everything in Build waits for everything in Design — including the drawing that has nothing to do with the foundation. The plan is now later than the work requires, and it is later for a reason nobody can point to, because the real dependency (foundation needs the layout, walls need the drawings) was never written down.
Three consequences follow, and they get worse as the plan grows:
- The critical path runs through containers. It tells you “Design is critical”, which nobody can act on, instead of “the drawings are critical”, which somebody can.
- Dates distort. When one detail task in Design slips, the whole of Build moves, even the parts that did not depend on it. The slip is real; the size of it is invented by the link.
- Change becomes dangerous. Insert a task into Design and it silently becomes a predecessor of the whole of Build. Most people do not notice until the finish date has moved and nobody knows why.
The rule is simple and admits no exceptions I have found in twenty years of looking at other people’s plans: summary tasks carry no links. If you feel the need to link two summaries, find the two detail tasks that actually depend on each other and link those. The summaries will follow, because that is what they are for.
Reading the links in Microsoft Project
Everything above lives in one column. The Predecessors field on each task lists the tasks it depends on, with the link type and the lag: 12 means finish-to-start from task 12 with no lag (FS is the default and is not written); 12SS+1d a start-to-start with one day of lag; 12FS-2d a finish-to-start with two days of lead. The Successors field shows the same links from the other side, and the Task Information dialog has a Predecessors tab that lays the same information out as a small table.
Two habits are worth keeping. First, read the Predecessors column rather than the arrows in the Gantt: on a plan of any size the arrows overlap and lie by omission, while the column is exact. Second, before a review, sort or filter for tasks with no predecessors and tasks with no successors: apart from the first and last milestones, every task should have both, and the ones that do not are where somebody typed a date instead of drawing a link. That check is one of the five checks worth running on any plan you did not build yourself.
Two things we measured in Project
Two behaviours of Microsoft Project are worth seeing on a real plan before they surprise you on yours. We measured both in Project 16.0.
- Start-to-start holds the start, not the finish. Plastering five days, painting five days, linked
1SS+2d: painting starts on Wednesday the 4th and finishes on Tuesday the 10th, plastering finishes on Friday the 6th. Then plastering stretches to eight days and finishes on Wednesday the 11th. Painting stays where it was and still finishes on Tuesday the 10th: the plan paints a wall that isn’t finished, and Project flags nothing, because the link is respected. A1FF+1dlink holds the finish (painting moves to Thursday the 12th) but leaves the start free. - Only one link between the same two tasks. Typing
1SS+2d;1FF+1din the painting’s Predecessors column, Project kept only the first link. Adding a second one through automation replaced the first. Either way the plan still had one link.
When you need the start and the finish tied at the same time, split the successor: “Painting, first coat” with SS+2d from plastering, “Painting, finish” with FS from plastering and from the first coat. Two real tasks, two durations you can estimate, and a plan that says what actually happens on site.
Where dependencies go wrong
The technique is old and the tools are good; the failures are all human, and they repeat.
- Links drawn to hit dates. The bar was dragged until it landed on the date in the contract. The link now models the contract, not the work, and the first slip breaks it.
- Every link treated as hard. When the finish does not fit, the only lever left is scope. Label the soft ones and you get a second lever for free.
- Buffers hidden in lags. “+5d to be safe” on twelve links is two months of float nobody can see or manage.
- Constraints instead of links. A typed date (“must start on”) is not a dependency; it is a promise the engine will keep even when the work behind it has moved. Almost every constraint in a plan is a link that was not drawn.
- Dependencies that cross the planning window. If you plan in waves, the far work is coarse — but a far package that constrains near work still needs its link, coarse as it is. That is covered in Rolling wave planning.
Frequently asked questions
What are the four types of task dependencies?
Finish-to-start (B cannot begin until A is complete), start-to-start (B cannot begin until A begins), finish-to-finish (B cannot be completed until A is completed) and start-to-finish (B cannot be completed until A begins). Finish-to-start is the default and by far the commonest; start-to-finish is rare.
What is the difference between lead and lag?
Lag is a delay after the link is satisfied — three days of curing before the walls go up. Lead is an overlap — the walls start two days before the last pour is finished. In Microsoft Project lead is entered as a negative lag, so the two are the same field with opposite signs.
What is a mandatory dependency and what is a discretionary one?
A mandatory (hard) dependency is imposed by the nature of the work — physics, safety, contract — and cannot be removed to save time. A discretionary (soft) dependency is a preference of the team, sensible but optional, and it is the first place to look when the schedule needs compressing. External dependencies come from outside the project and behave like hard links you do not control.
Can I link summary tasks in Microsoft Project?
Project lets you, and you should not. A summary task takes its dates from the tasks inside it; a link on the summary makes everything in one phase wait for everything in the other, hides the real dependency and makes the critical path run through containers instead of work. Link the two detail tasks that actually depend on each other and let the summaries follow.
How do I tell a real dependency from a link drawn to hit a date?
Ask what B needs from A. If the answer is a thing — a deliverable, a decision, a state of the site — the link is real. If the answer is “that is when it should happen”, the link is a date in disguise, and it will break at the first slip. Replace it with the real dependency if there is one, or with a named milestone if the date is a genuine commitment.
Is every task supposed to have a predecessor?
Every task except the first milestone should have at least one predecessor, and every task except the last milestone at least one successor. A task with neither is floating: its dates are typed, not scheduled, and it will not move when the plan does. Filtering for tasks without predecessors or successors is the fastest check of a plan’s logic.
Can I put a start-to-start and a finish-to-finish link between the same two tasks?
No. Microsoft Project keeps one link per pair of tasks: the second one replaces the first or is discarded. If you need both, split the successor into two tasks, each with its own link.
Further reading on this blog
- How to review a Microsoft Project plan you did not build: five checks
- Rolling wave planning: detail the next wave, leave the far future coarse
- How to update a Microsoft Project plan: baseline, status date and progress
- Your late tasks, in three clicks
- Work Breakdown Structure: how to build one that survives the project
Disclosure: this is the blog of a company that makes project-control software. Nothing above needs our tools: the four link types, lead and lag and the Predecessors column are standard Microsoft Project, and the rule about summary tasks is a habit, not a feature.
#TaskDependencies #ProjectScheduling #MSProject #ProjectManagement #CriticalPath