As　the　gene　expression　profiling　data　being　with　the　characteristic　of　severe　multicollinearity,　small　samples,　and　high　dimension,　it　is　difficult　to　build　tumor　classification　model.　Partial　least　square　regression　was　applied　as　dimension　reduction　method　to　the　model　of　tumor　classification.　Respectively,　principal　components　are　extracted　from　five　gene　expression　profiling　data　sets:　Gastric,　C.vs.SC,　Colon,　Lung　and　Acute　Leukemia.　Then,　the　extracted　principal　components　are　used　to　classify　the　samples　combining　with　SVM　method.　The　results　showed　that　the　partial　least　square　regression　combining　with　SVM　can　be　used　not　only　in　two-class　problem,　but　also　in　multiclass　problem　reliably.　©2010　IEEE.