This paper investigates the optimal dynamic paths of trade protection imposed on infant industries during the process of joining a free trade agreement. The framework is based on the dynamic learning-by-doing model developed in Melitz (2005), where industries are experiencing dynamic externalities. In this framework, restricted-time protection is introduced as a realistic approach to correspond to the conditions of actual agreements. According to the computational analysis, in some feasible cases of optimal tariff paths may not follow a downward trend, as conventional wisdom would suggest. The results of the numerical simulation applied to the Vietnamese motorcycle industry support these findings.