For each prediction sample, the neighbors, yielding the 12 smallest sum of squares of differences to the test sample using the 140 peaks, were extracted from the whole calibration set and a local model built using these 12 chromatograms and related dioxin content. Prediction performance was thereafter validated for 10 fully independent samples. The performance of this model, yielded a correlation of 0.85 (r2) and a root mean square error of prediction of 2.3 pg PCDD/F TEQWHO/ g fat.