This thesis studies optimal reconstruction and its errors of the United States surface air temperature and precipitation since 1895 and extensively uses Empirical Orthogonal Functions (EOF). The reconstruction is based on multivariate linear regression analysis using EOFs as the design matrix. Two datasets, named TOB and F52 from the United States Historical Climatology Network (USHCN), are used to build the model. Subsequently, the time series of these two datasets and the one obtained from our reconstruction are compared for error estimation through cross comparison analysis. The reconstructed data is very close to the slightly adjusted TOB data and has a larger deviation from the fully adjusted F52 dataset. The mean squared error (MSE) of the reconstructed temperature was computed. A lower MSE for temperature was observed in the summer months than in the winter months.