Introduction The objective of this study was to evaluate the potential of exercises performed in a 2D video-capture virtual reality (VR) training environment to improve upper limb motor ability in stroke patients compared to those performed in conventional therapy. Methods A small sample randomized control trial, in an outpatient rehabilitation center with 12 patients with chronic stroke, aged 33–80 years, who were randomly allocated to video-capture VR therapy and conventional therapy groups. All patients participated in four clinical evaluation sessions (pre-test 1, pre-test 2, post-test, follow-up) and nine 45-minute intervention sessions over a 3-week period. Main outcomes assessed were Body Structure and Function (impairment: Fugl–Meyer Assessment [FMA]; Composite Spasticity Index [CSI]; Reaching Performance Scale for Stroke), Activity (Box and Blocks; Wolf Motor Function Test [WMFT]), and Participation (Motor Activity Log) levels of the International Classification of Functioning. Results Improvements occurred in both groups, but more patients in the VR group improved upper limb clinical impairment (FMA, CSI) and activity scores (WMFT) and improvements occurred earlier. Patients in the VR group also reported satisfaction with the novel treatment. Conclusions The modest advantage of VR over conventional training supports further investigation of the effect of video-capture VR or VR combined with conventional therapy in larger-scale randomized, more intense controlled studies.