In this study, we investigated picture (Experiments 1 and 2) and word (Experiments 3 and 4) processing using different tasks. In Experiments 1 and 3, easy and difficult conditional naming tasks were compared to a free naming task. In Experiments 2 and 4, easy and difficult conditional naming tasks were compared to easy and difficult manual forced-choice semantic decision tasks. For pictures, we showed that a difficult semantic categorization determined a cost for the conditional naming with respect to the free naming (Experiment 1). Also, we found that the difference in RTs between the easy and difficult conditional naming tasks was much smaller than the difference between the easy and difficult forced-choice semantic decision tasks (Experiment 2). For words, results showed that free reading was faster than easy conditional reading, which in turn was faster than difficult conditional reading (Experiment 3). An analogous pattern of results was obtained when the easy and difficult conditional reading tasks were compared to the easy and difficult forced-choice semantic decision tasks (Experiment 4). Globally, the results showed that whether a cost is observed or not depends upon the relative timing of the classification and name retrieval processes. A theoretical framework has been proposed.