Reliable and efficient order estimation. The main building blocks of identification systems for linear or quasi-linear systems have been extensively studied in the literature. However, the efficient integration of these blocks is not solved yet. One of the main problems is the automatic estimation of the model order. Current methods scan a candidate set of model orders to determine the optimum. These methods are rather time consuming, they may find a local optimum instead of the global one, and a priori knowledge on the possible model order is necessary. We propose a joint order and model estimation method using stochastic iterative search. Model evaluation. Embedded identification systems must provide evaluation capabilities to support both automatic model evaluation and high level report generation. Unfortunately model evaluation requires expert knowledge and skills, no algorithmic solution is known so far. Our goal is to create and automatic evaluation system by formalizing available expert knowledge.