Prior to the development of any formal model of representation (in UML, ER models or DFD models), it is necessary to gather facts about an organisation, its needs and the constraints in which it operates.  The key activity is to collect a mass of facts, which must then be filtered, sorted and classified, since many facts may not be relevant and those that are, will be mixed up in a variety of forms.  Once classified, the selected facts need to be verified with users for relevance and correctness.

Gather facts fully before trying to understand

Many facts are inter-related and it is therefore important to collect as many as possible, before trying to filter, classify and understand them.  Consider fact gathering as a form of brainstorming, in which you are trying to collect as much information as possible, even if you know some of it may not be relevant.

Ensure a clear understanding of the requirements and their priorities

Requirements will drive the remainder of the development process; therefore requirements must be correct and as complete as possible.  Furthermore, not all requirements are of the same importance and so each must be prioritised to determine the most critical requirements.

Ensure the system requirements are expressed in conceptual terms, not physical terms 

A conceptual view of a requirement is where the requirement is expressed in terms of what must be achieved, as opposed to how it might be achieved.  Requirements should provide direct benefit to those end-users using or being supported by the proposed system.  Requirements which are about supporting the current process or those involved within the system are not key requirements and relate to the physical manner in which the system is currently implemented.

Use a variety of information gathering techniques

Successful analysis will involve the use of a variety of information gathering techniques, including personal interviews.  Each information gathering technique (brainstorming, questionnaires, interviewing, documentation review etc.) has its advantages and disadvantages and need to be matched to the problem under consideration, the reliability of information sources, the risk of project failure, and the resources available.

Beware of natural language descriptions

Natural language is not good for expressing requirements and system specifications- it can be ambiguous and easy to misinterpret.