The choice of a self-administered mode of administration presupposes that respondents are able to read the questionnaire sufficiently well to respond as intended. The current paper describes the nature of reading problems found in cognitive testing of self-administered Census forms that were pretested with respondents among whom reading problems may be expected. The identification of reading problems and the effects of various kinds of misreadings are discussed. When respondents alter substantive nouns, meanings may or may not be substantially altered. Changes in "grammatical functors" (like negatives, plural markers or prepositions) often create large differences in meaning, since they carry logical connections between substantive elements. Respondents in these special populations may be unfamiliar with conventions governing form completion, like marking answers only within the provided blocks. Respondents may also be unfamiliar with conventions governing the relevance of particular information to the question asked or to the survey in general. This may make it difficult for them to correct their own misreadings or misunderstandings. We suggest that such phenomena be termed "forms literacy", and should be further investigated. Further research should address conditions under which highly literate individuals exhibit problems associated with forms literacy.