Previous research demonstrates that simultaneous training of novel sound contrasts in both perception and production can disrupt rather than enhance perceptual learning, indicating that although perception and production are assumed to be closely connected, these modalities may have a competitive relationship. In spite of this perceptual disruption, subjects trained in perception and production show gains in producing the distinction they were trained on, compared to perception-only training. The current study examines how subjects learn to produce a new sound contrast after training in perception or production. L1 Spanish speakers were trained on an unfamiliar Basque sibilant fricative-affricate contrast: /s̺a/–/ʃa/. Since learners’ productions of the contrast may not be identical to the way native speakers distinguish it, and rather than exploring a single phonetic dimension, we apply Linear Discriminant Analysis to acoustic measurements of subjects’ post-test productions to classify whether and how they distinguish the categories in a potentially multidimensional space. This classification model is then applied across conditions to compare production learning across training modes and examine how production learning relates to perceptual learning.