This paper presents a novel method for classifying human tumor samples using microarray gene expression data. The method involves dimension reduction using Partial Least Squares (PLS) followed by classification using Logistic Discrimination (LD) and Quadratic Discriminant Analysis (QDA). The study compares PLS with Principal Components Analysis (PCA) and demonstrates that PLS often performs better, especially in cases where PCA fails. The proposed methods were applied to five different microarray datasets involving various human tumor samples, including normal versus ovarian tumor, Acute Myeloid Leukemia (AML) versus Acute Lymphoblastic Leukemia (ALL), Diffuse Large B-cell Lymphoma (DLBCL) versus B-cell Chronic Lymphocytic Leukemia (BCLL), normal versus colon tumor, and Non-Small-Cell-Lung-Carcinoma (NSCLC) versus renal samples. The stability of the classification results was assessed using re-randomization studies. The results show that PLS-based methods outperform PCA in most cases, particularly when the number of genes (p) exceeds the number of samples (N). The study also highlights a condition where PCA fails to predict well relative to PLS, emphasizing the importance of using PLS for high-dimensional gene expression data. The methodology is implemented using standard statistical methods and is available in SAS. The paper concludes that PLS is a valuable tool for analyzing microarray gene expression data, particularly in distinguishing between different tumor types and predicting patient survival times based on gene expression patterns.This paper presents a novel method for classifying human tumor samples using microarray gene expression data. The method involves dimension reduction using Partial Least Squares (PLS) followed by classification using Logistic Discrimination (LD) and Quadratic Discriminant Analysis (QDA). The study compares PLS with Principal Components Analysis (PCA) and demonstrates that PLS often performs better, especially in cases where PCA fails. The proposed methods were applied to five different microarray datasets involving various human tumor samples, including normal versus ovarian tumor, Acute Myeloid Leukemia (AML) versus Acute Lymphoblastic Leukemia (ALL), Diffuse Large B-cell Lymphoma (DLBCL) versus B-cell Chronic Lymphocytic Leukemia (BCLL), normal versus colon tumor, and Non-Small-Cell-Lung-Carcinoma (NSCLC) versus renal samples. The stability of the classification results was assessed using re-randomization studies. The results show that PLS-based methods outperform PCA in most cases, particularly when the number of genes (p) exceeds the number of samples (N). The study also highlights a condition where PCA fails to predict well relative to PLS, emphasizing the importance of using PLS for high-dimensional gene expression data. The methodology is implemented using standard statistical methods and is available in SAS. The paper concludes that PLS is a valuable tool for analyzing microarray gene expression data, particularly in distinguishing between different tumor types and predicting patient survival times based on gene expression patterns.