People occasionally encounter information of which the structure bears divergent evaluative implications. For instance, when reading that a sunscreen protects against skin cancer, the relational meaning of the information (i.e., "protects against skin cancer") has positive evaluative implications for the sunscreen, whereas the co-occurrence (of "sunscreen" with "skin cancer") is negative. An important theoretical (and practical) issue is whether the co-occurrence information influences people's evaluations beyond the relational meaning of the information. This question has been recently investigated using task comparison procedures (comparing evaluative outcomes on different tasks) and process dissociation procedures (estimating relational and co-occurrence parameters within a given task). In this article, we report four experiments that examined this question by reducing interpretational ambiguities inherent in the two preceding approaches. This was achieved by using self-reported and mouse-tracking measures of ambivalence. We reasoned that when co-occurrence and relational information have divergent (rather than convergent) evaluative implications, more ambivalence should be found. We tested this prediction in experiential (Experiments 1 to 3) and instructed (Experiment 4) procedures. Higher self-reported ambivalence was found in divergent compared to convergent conditions in all experiments. Ambivalence, as estimated with mouse-tracking measures, was higher in divergent than in convergent conditions in the experiential experiments but not in the instructed experiment. Results are discussed with reference to single-process (propositional and episodic) and dual-process attitude learning models.