Abstract Semantic substitution errors (slips of the tongue) naturally occurring in Russian normal speech were analyzed for word frequency, word length, target-error cooccurrence strength, and word association norms. Target word frequencies were found to be significantly lower than error word frequencies; besides, there is a very significant positive correlation between target and error frequency values. Contrary to the view that the frequency effect is located at the stage of phonological encoding, the results suggest that frequency is coded at an earlier stage of lexical selection. Word length is a significant variable that determines the outcome of the error for non-cohyponym target-error pairs but not for cohyponym pairs. At the same time, cohyponym target-error pairs are characterized by much higher cooccurrence measures and stronger associative links compared to non-cohyponym pairs. Theoretical implications of these findings are discussed