The problem of robust observer synthesis for large scale systems subjected to nonlinear interaction functions and bounded uncertainties is considered. The scheme of the continuous observer that is the improvement of a known discontinuous variable-structure-type law, is based on certain functional properties of the uncertainty bound. The gain matrices of the observation scheme are in terms of the solutions of constrained Riccati equations. Solvable conditions of the constrained Riccati problem are investigated and, then, an efficient procedure for searching the desired solutions is established. Finally, the observation error is guaranteed to converge exponentially by the proposed nonlinear observer. An example is given to show the application of the proposed constrained Riccati approach.