The paper deals with the estimation of parameters from perspective images. The parameters concerned include pose of objects and intrinsic object parameters, like object dimensions, intrinsic angles, etc. Our aim was to implement a flexible, model-driven algorithm, giving an optimal accuracy of the result. An object, or a scene is modeled as a tree graph, with nodes representing object parts, leaves representing object features, and branches representing spatial relations between them. This representation method is suitable for a broad class of objects, and allows an efficient automatic computation of parameters. The approach for the parameter estimation is Bayesian. The object model, statistics of image acquisition and feature extraction, and an a priori probability distribution of parameters are used to compute the maximum a posteriori (MAP) estimate of the parameters, as well as the accuracy of this estimate. Some implementation remarks and an illustrative example are given.