This paper experimentally investigates the interdependence between market competition and endogenously emerging inter-firm collaboration. We restrict attention to arrangements resulting from bilateral collaboration agreements that typically characterize real world applications in which the activity concerned is a core activity of the partnering firms and risk sharing, contract enforcement and protection of proprietory knowledge are central issues. We rely on a baseline model by Goyal and Joshi (2003) which formalizes the strategic formation of collaborative networks between firms that are competing on the same product market. This model predicts strategically stable patterns of inter-firm collaboration which are empirically observed but have been ruled out in the previous theoretical literature. In a two-stage game, firms decide to form bilateral collaboration links, whose formation is costly but reduces marginal production costs, before they compete in quantity on the market. We report the results of a series of experiments. The first experiment is designed as a straightforward theory-test simulating a one-shot interaction. We manipulate the cost of link formation in different treatments. Our data almost perfectly match the predictions for both stages whenever the link formation costs are extreme and the predicted networks symmetric (empty or complete networks). In the case of intermediate link formation costs where the predicted networks are asymmetric, subjects rarely form asymmetric networks. When they do, observed and predicted quantities are less in accordance than for symmetric networks. Collusion cannot account for the observed behavior. In our second experiment we reject the conjecture that these findings are driven out by experience in a setting in which we increase the implemented number of repetitions of the two-stage game. Finally, in our third experiment we reduce the complexity of the setting by transforming the original two-stage game into a one-stage game where the formation of inter-firm networks directly determines firms’ payoffs. These are derived from assumed equilibrium market outputs on the here absent competition stage. In this case, observed networks coincide with the predicted ones indicating that experimental subjects’ limited capacity to foresee the outcomes of the market stage may be driving the earlier discrepancies.