The purpose of this study was to investigate intrusion errors for consonants in short‐term memory (STM) when perceived in varying signal‐to‐noise (S/N) ratio conditions. Stimuli were eighteen English consonants paired with the vowel /a/, and presented in groups of nine blocked in threes. Twenty‐seven phonetically trained college students were separated into three groups of nine subjects each. Subjects in each group heard the stimuli in either (1) 0 dB S/N, (2) −5 dB S/N, or (3) −10 dB S/N ratios (reMCL); transcribed their responses; covered them; and recalled items from STM. A paradigm described earlier [W. Wickelgren, J. Acoust. Soc. Am.39, 388–398 (1966)] was used for stimulus presentation and recording of results. Subjects' error data were submitted to INDSCAL and analyzed for distinctive features. Features were compared across noise conditions and to earlier studies. Data were also analyzed in terms of number correct.