This quasi-experimental study investigated the effects on 558 grades five and six students of three different teaching conditions: the "classical" model of conceptual change (for which cognitive conflict is considered as a precondition to the transformation of knowledge), the "prevalence" model of conceptual change (in which different conceptions can coexist, with one of them surpassing the others), and "repetition of traditional teaching" (that avoids cognitive conflicts and concentrates on the automatization of appropriate thought processes). These conditions were reduced to sequencing considerations, as classical model participants were first subjected to a possible cognitive conflict induced by a video, followed by another video about the targeted conceptions; "prevalence" model participants were subjected to the same videos but in the opposite chronological order; and "repetition" condition participants watched the "traditional teaching" video twice. Differences in accuracy and response times between our computerized and validated "sink/float" pretest and retest were analyzed. Results and interpretations confirm that cognitive conflicts are useful in teaching sequences that aim at producing conceptual changes. However, the major findings of this research suggest that such conflicts should not necessarily be triggered at the very beginning of teaching sequences, and therefore that the prevalence model might possibly be the preferable one to promote conceptual changes in real-life school science teaching settings. Recommendations for teaching and research are formulated. Presented results, although statistically significant, sometimes show weak effects sizes, and therefore call for further research efforts.