The deterioration of speech intelligibility in the presence of other sound sources has been explained in terms of both energetic masking, which renders parts of the speech signal inaudible, and informational masking, in which audible components of the masker interfere with speech identification. The current study focuses on the role of a specific form of informational masking in which audible glimpses of both target and masker combine to produce an incorrect listener percept. We examine a corpus of word misperceptions in Spanish which occur when target words are combined with a babble masker. Glimpses originating in both the target and the masker are force-aligned to the reported misperceived word in order to identify the most likely acoustic evidential basis for the confusion. In this way, the degree of involvement of both target and masker can be quantified. In nearly all cases, the best explanation for the misperception involves recruiting evidence from the babble masker (type I error), and in more than 80\% of the tokens some of the audible target evidence is ignored (type II error). These findings suggest misallocation of acoustic-phonetic material plays a significant role in the generation of speech-in-babble confusions.