A plethora of studies demonstrated that repeated selective retrieval of target items from semantic categories has an adverse memory effect on semantically-related memories, a phenomenon called retrieval-induced forgetting (RIF). However, there is a range of boundary conditions for RIF. For instance, forming interconnections between target and competitors, long-term delay without sleep between practice and final recall, and the form of learning all attenuate the effect of selective practice on the accessibility of semantically-related competitors. The aim of the present research was to investigate the latent general preconditions behind the reductions of RIF. In Experiment 1 participants learned category-exemplar pairs with repeated study or combined study-full test sessions followed by a selective retrieval practice and a full cued-recall test. We found lower difference between performance on the non-practiced items from practiced categories (Rp-) and their baseline from unpracticed categories (Nrp-) suggesting a reduction in retrieval-induced forgetting. However, regression analysis revealed that this was possibly caused by the increased recall performance independently from the presence of an initial full retrieval session. Therefore, in Experiment 2 participants learned the same pairs through a study followed by two study, two test or combined study-test cycles. As the consequence of increased rate of learning we confirmed the complete absence of retrieval-induced forgetting with Bayes factor analysis. Our results suggest that the adverse memory effect of selective retrieval practice shows a non monotonic dependency on the strength of the mnemonic representations.