We studied the expression profiles of various stages of colorectal tumors (adenoma (AD), seven samples; carcinoma (CA), 16 samples) by using cDNA microarrays and developed ADMS (algorithm for diagnosing malignant state) method, selecting 335 clones characteristic of CA state. We, then, applied ADMS to 12 additional samples (five from primary lesions with metastasis and seven metastases); all 16 CAs and 12 metastatic tumors were diagnosed correctly as cancerous states. Although three of the seven ADs were diagnosed as "cancerous," the large size of two of these tumors suggested their potential malignancy. Our strategy for selecting clones characteristic of the malignant state is widely applicable to diagnosis and for predicting the stage of progression during multistep carcinogenesis. Of the 335 clones we selected, 135 were known genes. Included in the 135 genes were tumor suppressor and growth factor-related genes and were consistent with the literature. ADMS is a reliable means for identifying genes useful for the diagnosis of cancer.