AbstractThe problem of parameter estimation from bounded‐error data is considered. In particular, the possibilities of using an l2projection (least squares) procedure are explored. Exact as well as approximate (polytopic) outer‐bounding solutions are proposed. the main properties and difficulties in implementation of the l2projection procedures are discussed and illustrated. Since these methods appear to be computationally cumbersome when the number of measurements is large, a simpler (rectangular) characterization of the solution set, as a result of natural interval calculations, is proposed and discus