Speech errors are a critical source of data on the tacit knowledge that underlies our creative use of language. Studies of errors in spontaneous speech, in experimental paradigms such as tongue twisters, and those produced by aphasic individuals reveal the influence of linguistic principles on the production of speech. Linguistic representations from distinctive features and gestures to prosodic structure shape speech error distributions. Speech errors respect allophonic and allomorphic relationships between linguistic forms. Phonological well-formedness – as measured by cross-linguistic markedness and within-language frequency – influences which structures are (mis)produced in errors. Recent advances in constraint-based generative phonological theories (Optimality Theory, Harmonic Grammar, and Maximum Entropy Grammars) may provide insight into these patterns. Formal models of variation and computational models of constraint satisfaction may provide the means to quantitatively link phonological principles and speech error distributions.