Several markers of malignancy have been proposed for canine mammary tumors on the mRNA and protein levels. However, their association with tumor malignancy applies only for mean values of large groups of tumors, but no single marker identified to date can be used to reliably predict malignancy for individual tumors. A quantitative real-time reverse transcription polymerase chain reaction array was established to quantify the expression levels of 49 genes relevant to carcinogenesis in laser-microdissected tumor cells of 10 benign and 13 metastatic canine mammary tumors. Analysis of variance and discriminant analysis were used to identify relevant gene expression patterns that differentiate adenomas from metastatic carcinomas and their lymph node metastases. Seventeen genes with significant (P < .05) differences in gene expression levels between benign and malignant tumors were identified--including ERBB1, SLIT2, progesterone receptor, MIG6, SATB1, and SMAD6--but correct classification of each tumor as benign or malignant was impossible on the basis of any of these genes alone. However, the combined expression patterns of BMP2, LTBP4, and DERL1 (Derlin-1) correctly classified each individual tumor as benign or malignant. This pilot study identified a complex mRNA expression pattern of 3 genes that was able to identify malignancy in laser-microdissected tumor cells for each individual tumor, instead of group means as used in previous studies.