Abstract The study of recollective and nonrecollective retrieval has become controversial, owing to several critiques of traditional recognition-based measurement (e.g., remember/know, ROC, process dissociation). We present a new methodology in which subjects merely study and recall lists, using any standard paradigm (associative, cued, free, or serial recall), the data are analyzed with a Markov model whose parameters measure recollective and nonrecollective retrieval, and the model’s fit is compared to that of one-process models. The power of this approach is illustrated in some experiments that dealt with two classic questions: (a) What are the process-level differences between associative and free recall, and (b) why does taxonomic organization improve free recall but impair associative recall? Fit results showed that a dual-retrieval model is both necessary and sufficient to account for associative and free recall data, in contrast to the sufficient-but-not-necessary pattern that prevails in the recognition literature. Key substantive findings were that associative recall is more reliant on recollective retrieval and less reliant on nonrecollective retrieval than free recall, that taxonomic organization impairs recollective retrieval in both paradigms, and that taxonomic organization enhances the reconstruction component of nonrecollective retrieval in free recall.