The knowledge of induction generator models and their parameters has gained great importance in recent years. Induction generators have been widely used in several applications, including renewable energy systems, because of their simple construction and easy operation. A successful parameter's estimation of induction generators strongly depends on the availability of a good initial parameter guess. When it is not available, the estimation process could take plenty of time to converge or even to diverge. This paper proposes a hybrid method that estimates parameters of induction generator transient models from disturbance measurements through a hybrid algorithm based on trajectory sensitivity and mean-variance mapping optimization. The method is robust regarding initial parameter guesses, requires no disconnection of the generator from the grid, and uses measurements commonly available in practice, such as generator terminal voltage and current. The system modeling for estimation purposes is based on a squirrel-cage induction generator, represented by a third-order model, connected to both a grid and a static load. The method was tested with actual measurements obtained from a small sized power system designed in the laboratory. The results show correct estimates were successfully achieved and the model can represent the dynamic response of the system accurately.