This is a review of published theory and research on the assumptions that underlie Miller's conflict models. Operational distinctions are made between single-goal and multiple-goal conditions and, within the latter, the spatial, temporal, and stimulus dimensions. Major operational divergences in research outcomes are single- vs. multiple-goal conditions and strength-of-pull vs. other measures. Support for Miller's postulate system is equivocal, with considerable support available for several alternate postulates. A preferred interpretation is that response gradients can assume any degree of steepness or generality as a function of several manipulation and selection conditions. One manipulation variable, time before testing, produces flatter gradients which simulate unpracticed time-course effects. A major unfinished research area deals with the summation hypothesis or, more generally, specification of principles of combination for response gradients.