Optimal design of high-speed valve trains requires the use of an accurate analytical model. While the governing differential equations are important, the coefficients (or parameters) used in these equations are equally as important. Since many of the parameters used in valve train models are difficult to measure directly, parameter identification based on experimental data is required to assure model accuracy.This paper addresses the parameter identification problem for a valve train model, formulating a scalar cost function which represents the difference in measured and predicted system response. Minimization of this cost function yields the 10 unknown system parameters. As the cost function has many local minima, a global optimization scheme must be employed. An implicit filtering algorithm is implemented which applies a scale reduction scheme in conjunction with a gradient projection algorithm to avoid becoming trapped in local minima and thus produces near global minima of the cost function. The implicit filtering algorithm has several tuning parameters which allows its adaptation to many problems. For this problem, implicit filtering proved to be 2 to 4 times more efficient than the adaptive random search method previously employed.