Rats were trained with two Pavlovian serial feature positive discriminations (X→A+/A- and Y→B+, B-), in which the target stimuli (A and B) were reinforced when they were preceded by the feature stimuli (X and Y), but not when they were presented alone. In one discrimination the feature-target interval was 10 s, and in the other discrimination that interval was 30 s. In a series of tests in which the feature target interval varied in ten steps between 5 and 50 s, performance was better when the targets were presented at their usual training times after the feature, rather than earlier or later. Training with the shorter FTI (10 s) resulted in both greater peak responding at the training interval and steeper temporal gradients than training with the longer FTI (30 s). Presentation of a compound feature comprised of both the 10 s and 30 s feature elements shifted the time of peak target responding to an intermediate value, 15 s. In contrast, simple conditioning controlled by the features (measured in the empty intervals between feature and target presentations) showed temporal control but no evidence for stimulus averaging. Most of these results are consistent with scalar timing, but may imply separate temporal codes or timing mechanisms for simple conditioning and occasion setting.