This paper addresses the worst-case parameter identification problem for uncertain single-input/single-output (SISO) and multi-input/multi-output (MIMO) linear systems under partial state measurements and derives worst-case identifiers using the cost-to-come function method, In the SISO case, the worst-case identifier obtained subsumes the Kreisselmeier observer as part of its structure with parameters set at some optimal values, Its structure is different from the common least-squares (LS) identifier, however, in the sense that there is additional dynamics for the state estimate, coupled with the dynamics of the parameter estimate in a nontrivial way, In the MIMO case as well, the worst-case identifier has additional dynamics for the state estimate which do not appear in the conventional LS-based schemes, Also for both SISO and MIMO problems, approximate identifiers are obtained which are numerically much better conditioned when the disturbances in the measurement equations are "small." The theoretical results are then illustrated on an extensive numerical example to demonstrate the effectiveness of the identification schemes developed.