Three experiments were designed to test whether experimentally created ad hoc associative networks evoke false memories. We used the DRM (Deese, Roediger, McDermott) paradigm with lists of ad hoc categories composed of exemplars aggregated toward specific goals (e.g., going for a picnic) that do not share any consistent set of features. Experiment 1 revealed considerable levels of false recognitions of critical words from ad hoc categories. False recognitions occurred even when the lists were presented without an organizing theme (i.e., the category’s label). Experiments 1 and 2 tested whether (a) the ease of identifying the categories’ themes, and (b) the lists’ backward associative strength could be driving the effect. List identifiability did not correlate with false recognition, and the effect remained even when backward associative strength was controlled for. Experiment 3 manipulated the distractor items in the recognition task to address the hypothesis that the salience of unrelated items could be facilitating the occurrence of the phenomenon. The effect remained when controlling for this source of facilitation. These results have implications for assumptions made by theories of false memories, namely the preexistence of associations in the activation-monitoring framework and the central role of gist extraction in fuzzy-trace theory, while providing evidence of the occurrence of false memories for more dynamic and context-dependent knowledge structures.