Conversational speech is the most socially valid context for evaluating speech intelligibility, but it is not routinely examined. This may be because it is difficult to reliably count the number of words in the unintelligible portions of the sample. In this study four different approaches to dealing with this problem are examined. Each is based on the assumption that it is possible to perceive syllables in unintelligible strings even when the target words are unknown; these unintelligible syllables can then be used to estimate the number of unintelligible words in these samples using at least four different approaches. Preliminary data are presented for each of the four approaches based on conversational speech from two convenience samples including 320 children with normal (or normalized) speech and 202 children with speech delay. Differences among the four approaches are discussed. (Contains 4 tables.)