This paper presents a theory for convergent state estimators when the design model has parameter errors, truncated modes, unmodeled disturbances, and neglected nonlinearities of a special class. The tolerable class of nonlinearities and disturbances is that which yields square-integrable model error vectors. The potential advantage of the method is that the total model error vector is estimated, rather than just the parameter estimation (identification) presently available with parameter adaptive methods. The proposed method uses the orthogonal filter in an adaptive manner to get improved estimates of the model error vector simultaneously with estimates of the reduced model state vector. © 1981 American Institute of Aeronautics and Astronautics, Inc., All rights reserved.