This paper presents a parameter identification technique for multibody dynamic systems, based on a nonlinear least-square optimization procedure. The procedure identifies unknown parameters in the differential-algebraic multibody system model by matching the acceleration time history of a point of interest with given data. Derivative information for the optimization process is obtained through dynamic sensitivity analysis. Direct differentiation methods are used to perform the sensitivity analysis. Examples of the procedure are presented, applying the technique both to perfect data; i.e. data produced by the assumed model with the optimal choice of parameters, and to experimental data; i.e. data measured on the real system and thus subject to noise and modelling imperfections.