The aim of this paper is to provide firm-level evidence on the short-run link between outsourcing and labor productivity using an original dataset of Italian manufacturing firms, and applying a two-stage probit least squares estimator. We find a positive effect on productivity from outsourcing only if firms provide training for the workforce. This indirect impact on productivity is independent of the type of activity outsourced and is bigger in the case of service outsourcing. This can be explained by the different feedback effect of labor productivity on training and by the different type of training provided. While production outsourcing induces an organizational change which stimulates off-the-job training for plant operators, service outsourcing induces firms to train a broader range of occupational profiles - both off and on the job. Similar results emerge for the case of joint outsourcing of both production and service activities. Therefore, we find that outsourcing generates positive productivity effects only if it is part of a broader knowledge management strategy that involves upgrading of workers’ skills.