Accurate crack propagation simulation requires critical fracture parameters to be known a priori. For elastic-plastic materials, two fundamental parameters are the separation energy and the peak stress required to generate new crack surfaces. In general, both are difficult to quantify since direct determinations are not possible in experiments. For inhomogeneous materials, such as graded materials, determination is even more complex since these parameters vary spatially. In this paper, a novel method based on an inverse analysis technique is proposed to estimate the fracture parameters of elastic-plastic and graded media. The method utilizes the Kalman filter to process measured data and extract best estimates of the unknown parameters. The accuracy of the method is examined in a verification study where a dynamically propagating crack in double cantilever beam type specimen is modeled. In the study, time variation records of crack opening displacement, opening strain, crack advance distance, and load point reaction force are used as possible measurements. Despite large noises in data, the results confirm accurate estimation. The estimates improve when multiple measurements are supplied to the inverse technique.