Many studies into learners' ideas in science have reported that aspects of learners' thinking can be represented in terms of entities described in such terms as alternative conceptions or conceptual frameworks, which are considered to describe relatively stable aspects of conceptual knowledge that are represented in the learner's memory and accessed in certain contexts. Other researchers have suggested that learners' ideas elicited in research are often better understood as labile constructions formed in response to probes and generated from more elementary conceptual resources (e.g. phenomenological primitives or "p-prims"). This "knowledge-in-pieces perspective" (largely developed from studies of student thinking about physics topics), and the "alternative conceptions perspective," suggests different pedagogic approaches. The present paper discusses issues raised by this area of work. Firstly, a model of cognition is considered within which the "knowledge-in-pieces" and "alternative conceptions" perspectives co-exist. Secondly, this model is explored in terms of whether such a synthesis could offer fruitful insights by considering some candidate p-prims from chemistry education. Finally, areas for developing testable predictions are outlined, to show how such a model can be a "refutable variant" of a progressive research programme in learning science. (Contains 2 figures.)