Non-exact replications are regarded as effective tools of problem solving in psycholinguistic research because they lead to more plausible experimental results; however, they are also ineffective tools of problem solving because they trigger cumulative contradictions among different replications of an experiment. This paper intends to resolve this paradox by putting forward a metatheoretical model that clarifies the criteria with the help of which various aspects of the effectiveness of the problem solving process can be differentiated and evaluated. The key point is the reconstruction of the relationship between original experiments and their non-exact replications by introducing the concept of "experimental complex" and analysing the problem solving strategies that the researchers apply. The workability of the proposed metatheoretical model is illustrated with the help of three case studies.