Impaired procedural learning has been suggested as a possible cause of developmental dyslexia (DD) and developmental language disorder (DLD). We evaluate this theory by performing a series of meta-analyses on evidence from the six procedural learning tasks that have most commonly been used to test this theory: the serial reaction time, Hebb learning, artificial grammar and statistical learning, weather prediction, and contextual cuing tasks. Studies using serial reaction time and Hebb learning tasks yielded small group deficits in comparisons between language impaired and typically developing controls (g = -0.30 and -0.32, respectively). However, a meta-analysis of correlational studies showed that the serial reaction time task was not a reliable correlate of language-related ability in unselected samples (r = 0.03). Larger group deficits were, however, found in studies using artificial grammar and statistical learning tasks (g = -0.48) and the weather prediction task (g = -0.63). Possible reasons for the discrepancy in results from different tasks that all purportedly measure procedural learning are highlighted. We conclude that current data do not provide an adequate test of the theory that a generalized procedural learning deficit is a causal risk factor for developmental dyslexia or developmental language disorder.