Purpose: Error variability has traditionally been considered a hallmark of apraxia of speech (AOS). However, in some of the current AOS literature, relatively invariable error patterns are claimed as a mandatory criterion for a diagnosis of AOS. This paradigm shift has far-reaching consequences for our understanding of the disorder and for its (differential) diagnosis. Against the background of this controversy, the present article aims to further examine error variability in AOS. Method: Four patients with relatively pure, mild-to-moderate AOS participated in the study. They repeated 8 target words in 2 different phrase contexts, 10 times each. Error analyses were based on phonetic transcription. Error variability was determined using several measures of (a) consistency of error occurrence and (b) consistency of error type. Results: All patients produced highly inconsistent reactions across multiple trials in some of the target words. However, other words were more consistently accurate or inaccurate. Several factors influencing error variability were identified. Conclusions: Because this study has disclosed clear indications of variable behavior in AOS, diagnostic guidelines claiming error consistency as a mandatory criterion cannot be maintained. Because error variability is difficult to operationalize, we recommend to no longer use (in)consistency as a strict diagnostic marker of AOS.