Conductive elastomers are a novel strain sensing technology which can be unobtrusively embedded into a garment's fabric, allowing a new type of sensorized cloths for motion analysis. A possible application for this technology is remote monitoring and control of motor rehabilitation exercises. The present work describes a sensorized shirt for upper limb posture recognition. Supervised learning techniques have been employed to compare classification models for the analysis of strains, simultaneously measured at multiple points of the shirt. The instantaneous position of the limb was classified into a finite set of predefined postures, and the movement was decomposed in an ordered sequence of discrete states. The amount of information given by the observation of each sensor during the execution of a specific exercise was quantitatively estimated by computing the information gain for each sensor, which in turn allows the data-driven optimization of the garment. Real-time feedback on exercise progress can also be provided by reconstructing the sequence of consecutive positions assumed by the limb.