For all systems it is useful to make at least two levels — the context diagram and the level one diagram. In fact, when in the earlier description of how to create DFDs you were told to start by identifying the external entities and then to identify the inputs and outputs of the system, you were learning how to produce the context diagram. The rest of the description was how to produce the level one diagram.

Whenever you perform data-flow modeling, start in exactly this way, producing a context diagram and then a level one diagram. Of course, in producing the level one diagram you may realise you need more inputs and outputs and possibly even more external entities. In this case, simply add the new data-flows and the new entities to the level one diagram and then go back and add them to the context diagram so that both diagrams still balance. Conversely, you may realise that some of the inputs and outputs you originally identified are not relevant to the system. Remove them from the level one diagram and then go back to the context diagram and make it balance by removing the same inputs and outputs.

This constant balancing between diagrams is very common when doing levelling.

What about making more levels? There are two reasons for making more levels. The first is the obvious one: you, as the software engineer, have not fully described a process to your satisfaction, so you expand that process into a next level diagram. The new diagram is built in just the same way that a level one is built from a context diagram only the new inputs and outputs are precisely to the data flows to and from the process you are expanding.

The second reason is that you realise the diagram you are working on is becoming cluttered and unclear. To simplify the diagram, collect together a few of the processes. Ideally, these processes should be related in some way. Replace them with a single process and treat the original collection of processes as a lower level, expanding the new process. The inputs and outputs to the new process are whatever inputs and outputs that are needed to make the diagrams balance. Remember to re-number the old processes to show that they have been moved down a level.

When doing this, if there is a data-store which interacts with these processes, and only these processes, then this too can be put on the lower level diagram.

Do not group random processes together to make a lower level diagram. This will only end up in a tangle of arrows and unrelated processes. A good guide as to whether or not you have chosen a sensible collection is try coming up with a new name for the replacement process. If you cannot do this then you have probably made too general a grouping. Perhaps leave out one or two processes or try a different grouping.

Always bear in mind that levelling is meant to simplify and clarify the diagrams, and if this cannot be done then it may be best to leave the diagram as it is.