This paper suggests a method of estimation, that is, the least squares estimator in the general situation, when observations are interdependent or independent. The method is designated as Perfect Least Squares (PLS) because there is no other method, known so far, that provides lower magnitude of the optimality criterion. The method holds good for data collected according to any sampling or census method. It is shown in this paper, empirically as well as theoretically, that PLS estimator scores over OLS and GLS estimators. The method is also extended to simultaneous equation systems. It can be applied straightaway to dynamic models.