Four experiments established complex choice responses a few seconds in duration. A response was reinforced if it sufficiently approximated a `target' response. The two target responses in each experiment were linear in the sense that they involved either constant rates, or constant rates of change, in component key pecking by pigeons. For example, in Experiment 4, one target response consisted of the linearly increasing pattern of 0, 1, 2, and 3 pecks per s in four successive seconds, and the other response consisted of 3, 2, 1, and 0 pecks per s. Contingencies were `tolerant' in the sense that they permitted variability across different reinforced exemplars of a response. Responses approximated target responses, at least crudely in terms of overall cumulative records, and sometimes even quantitatively in terms of within-trial, local sequential organization. In this sense, the `contents' of at least some choice responses, like their `envelopes', can be shaped. That is, patterning within their boundaries, in addition to their relative and absolute durations, can be shaped. Some responses may have emerged from variability inherent in component pecking occurring at a constant probability, thereby demonstrating a few cases where the relation between molecular and molar analyses fully legitimizes a molar analysis.