Colorectal-carcinoma specimens are heterogeneous and include areas of nonmalignant mucosal and connective tissue. For those study designs in which laser microdissection and RNA preamplification are impracticable, the optimal yield of genuine cancer RNA is a key factor in gene-expression analysis. In this study we compared alternative methods of tissue purification. Three contiguous 0.5-cm(3) samples taken from an advanced primary adenocarcinoma of the sigmoid colon were processed immediately after surgery with the use of the following methods: (1) cryotomy after manual dissection (CMD), (2) microscopically assisted manual dissection (MAMD), and (3) tumor-cell isolation with the use of Ber-EP4 antibodies and Dynabeads (Dynal Biotech GmbH, Hamburg, Germany; technique abbreviated as DB). We generated gene-expression profiles with the use of GeneChip technology (Affymetrix, Santa Clara, Calif) and recorded preparation times, costs, and RNA quantity and quality. CMD took 60 minutes, MAMD 180 minutes, and DB 90 minutes to isolate 22, 8, and 23 microg of RNA, respectively. Expenses for materials amounted to 41, 23, and 91 US dollars for CMD, MAMD, and DB, respectively. The 3'/5' ratio, as determined with the GeneChips, for GAPDH/beta-actin was 1.01:1.03 for CMD, 1.13:1.28 for MAMD, 1.43:1.68 for DB, K-ras, APC, smad 2, transforming growth factor-beta, and p53 were marked as present in all cases, with the exception of APC, which was graded as marginal on DB. The correlation values of gene-expression profiles were 91% (CMD/DB), 93% (CMD/MAMD), and 97% (DB/MAMD). All 3 methods provided enough RNA, of sufficient quality, for gene-expression microarray analysis in colorectal carcinoma. Cross-methodologic analyses of array data should not be performed uncritically.