An extraction method of intra-die spatial correlation using spectral density is presented. Based on theoretical analysis of random field, we select a widely-used parametric correlation function form whose unknown parameters are estimated effectively in the spectral domain. Compared with the existing extraction algorithm in the original spatial domain, our method can obtain same quality of results, while a sampling scheme closer to the reality is adopted. In actual measurement process, the unavoidable measurement error with unpredictable amount and type would greatly confound the extraction results. Our method can detect useful information from the error noise contamination through filtering technique, and provides feedback about the correctness of the measured data. Experimental results show that our method is more practical and stable.