One of the major findings of traditional investigations of speech errors is that a single segment is the smallest unit affected by speech errors, although the possibility that units smaller than the segment could play a role in explaining speech errors was not entirely rejected. A number of relatively recent studies using instrumental kinematic techniques for speech analysis bring evidence that errors often occur at subsegmental units of speech production, i.e. at the level of articulatory gestures. Such errors occur due to the coproduction of articulatory gestures, whereby the gestures from both the target consonant and the competing speech sound overlap. As they represent the gradient shift from one segment to another, they are often called gradient errors. Such processes are almost impossible to capture without the use of instrumental kinematic techniques, such as electropalatography (EPG). Th ere are no instrumental kinematic studies of speech errors in Croatian speech. Th us, the aim of this paper is to use EPG to investigate speech errors produced in one Croatian tongue twister. The analysis was focused on /r/ and /l/ targets produced by 10 native female speakers of Croatian, while producing the tongue twister which facilitated speech errors in these two sounds. Each token of the target consonant was classifi ed in one of the four categories: (1) perceptually and articulatorily correct production (P1A1); (2) articulatorily correct but perceptually incorrect production (P0A1); (3) perceptually correct but articulatorily incorrect production (P1A0) and (4) perceptually and articulatorily incorrect production (P0A0). Th e classifi cation was made by the authors via auditory analysis and visual inspection of spectrograms and electropalatograms prior to the quantitative kinematic analysis. Subsequent analyses showed evidence of gradient errors, which would not be detected without the use of instrumental kinematic techniques. Th is investigation supports the claim that traditional method of collecting speech errors by perceptual analysis only is not sensitive enough to detect the subtleties of erroneous productions and speech motor control.