Regression-type techniques are widely used for system modeling and characterization. In many of the cases the characterizations are to be performed on-line to be able to support control actions and other decisions which are necessary for the operation. In autonomous time critical and embedded systems there are further requirements to be met. Robustness and flexibility in respect to the actual state of the system and its environment belong to this group because the available time, resource, and data conditions have a direct effect on the feasibility and quality of the modeling or characterization. An important expectation concerning the processing is to ensure continuous operation and to offer "immediate" results in certain (e.g. crisis) situations. Anytime tools are serious candidates to measure up to such purposes because they can always provide some kind of results even if abrupt changes, temporal shortage of computational power, and/or loss of some data occur in the system/environment. In this paper an anytime model regression technique is presented which can advantageously contribute to the modeling/characterization tasks.