Three theoretical measures of Prägnanz were compared with four data sets. The theoretical measures were a stimulus-coding one (structural information load, SIL), a measure related to within memory processes (stability), and one based on the interaction of perception and memory (resonance). The four data sets were obtained in two experiments and involved goodness rating, grouping, and immediate and delayed recall. A complete set of seven-element binary serial patterns was used in each experiment. Both SIL and resonance were shown to correlate reliably with the data sets across tasks. The resonance measure, however, performed best. Prägnanz thus appears to be explained better by resonance than by stimulus coding or memory storage. Resonance explained all systematic variance in the recall tasks, but not in the other tasks. Regarding these, partial-correlation analyses showed that the effect of stability could be fully reduced to resonance. SIL could not be similarly reduced. Therefore, additional perceptual constraints, other than resonance, would be needed for a complete account of goodness in the judging or grouping tasks.