Sequences of behavior are highly predictable (stereotyped) during some segments and less predictable during transitions between those segments. Statistical characterization of behavior must involve observation of the behavior under different stimulus conditions, which includes how stimuli associated with behavioral releasers come to trigger behavior during learning. For example, the feeding motor program (FMP) of the honey bee Apis mellifera (Hymenoptera: Apidae) during proboscis extension can be divided into three response phases (Rehder 1987). We conditioned honey bees in an olfactory conditioning paradigm with one or several rewarded trials during which an odorant was paired with the sugar-water unconditioned stimulus (US); the latter elicits proboscis extension and feeding in properly motivated bees. By recording electromyogram activity from one of the muscles that move the proboscis during feeding, we quantified the bees' responses during an unrewarded test with either the conditioned odorant, a different (novel) odorant, or the sugar-water US. Various parameters of the response phases of the FMP varied in a consistent manner across these experimental treatments, with certain stimuli eliciting stronger, more consistent responses. The different response phases followed one another in time with some variability and statistical uncertainty. For example, the length of an individual licking movement with the glossa was relatively invariant, and may indicate that this parameter can be used to differentiate the FMP into more basic, independent units. Our work shows how learned information may release action patterns in ways slightly different from traditional sign-stimuli releasers.