Definitions of dyslexia typically make reference to unexpected poor reading, although how best to operationalize unexpected remains an issue. When operationally defined as reading below expectations based on level of oral language, cases of unexpected poor reading make up fewer than half of cases of poor reading, and cases of unexpected poor reading occur throughout the range of reading proficiency. An implication is that what optimally predicts poor reading may not optimally predict unexpected poor reading. The goal of the three presented studies was to test this implication empirically. In Study 1, a model-based meta-analysis, phonological awareness accounted for 40% of the variance in decoding but only 1% of the variance in decoding that was unexpected based on level of vocabulary. Conversely, unexpected phonological awareness accounted for 34% of the variance in unexpected decoding but only 1% of the variance in decoding. An analogous pattern of results occurred for reading comprehension. In Study 2, a study of 766 children in kindergarten, first grade, and second grade, latent variables were used to represent oral vocabulary, phonological awareness, and decoding. As was seen in Study 1, unexpected decoding was better predicted by unexpected phonological awareness than by phonological awareness. In Study 3, a longitudinal study of 1,025 children followed from preschool through grade 2, the pattern of results mirrored those of Studies 1 and 2. An important implication of these studies is that typical assessments may be better at identifying poor reading than they are at identifying unexpected poor reading or dyslexia.