A spectral domain method for intra-die spatial correlation function extraction is presented. Based on theoretical analysis of random field, the spectral density, as the spectral domain counterpart of correlation function, is employed to estimate the parameters of the correlation function effectively in the spectral domain. Compared with the existing extraction algorithm in the original spatial domain, the proposed method can obtain the same quality of results in the spectral domain. In actual measurement process, the unavoidable measurement error with arbitrary frequencycomponents would greatly confound the extraction results. A filtering technique is further proposed to diminish the high frequency components of the measurement error and recover the data from noise contamination for parameter estimation. Experimental results have shown that the proposed method is practical and stable.