Ditto DeeDee. Depends on the referral question.
In general, classroom observations should not involve the observer manipulating the environment. The observations can be recorded in a number of ways, though, such as a running record (the simplest form of observation), interval sampling (record the behavior occurring every 30", for example), frequency counts, observing specifically for antecedent-behavior-consequence behavioral sequences (usually focusing on predetermined target behaviors). If an experimenter is manipulating the environment, you usually need an additional observer to record the behavior (such as in an ADOS or ADOS-2 arena evaluation for autism).
And yes, in a naturalistic observation, ideally there should be comparison data of the same form taken on peers.