The strength based mirror effect (SBME) refers to an increase in hit rates (HR) and a decrease in false alarm rates (FAR) for the test lists that follow a strongly encoded study list. Earlier investigation of accuracy and reaction time distributions by fitting the diffusion model indicated a mirror effect in the drift rate parameter, which was interpreted as an indication of more conservative responses due to a shift in the drift criterion. Additionally, the starting point for the evidence accumulation was found to be more liberal for the strong test lists. In order to further investigate this paradoxical effect of list strength on these two kinds of bias estimated from the diffusion model, we employed the response-deadline procedure which provided a direct assessment of response bias early in retrieval, prior to evidence accumulation. Results from the retrieval functions indicated more liberal response bias in the list strength paradigm with both pure- and mixed-strength study lists. On the contrary, the SBME was observed at the asymptotic accuracy, suggesting that the conservative response bias might be observed later in retrieval when memory evidence has fully accumulated. In addition, comparison of the SBME across pure and mixed lists revealed that the SBME was most prominent in the pure-list paradigm, suggesting that both the differentiation and criterion shift accounts jointly explain the SBME in recognition memory. (C) 2014 Elsevier Inc. All rights reserved.