RNA expression analysis is an important tool in cancer research, but a limitation has been the requirement for high-quality RNA, generally derived from frozen samples. Such tumor sets are often small and lack clinical annotation, whereas formalin-fixed paraffin-embedded (FFPE) materials are abundant. Although RT-PCR-based methods from FFPE samples are finding clinical application, genome-wide microarray analysis has proven difficult. Here, we report expression profiling on RNA from 157 FFPE tumors. RNA was extracted from 2- to 8-year-old FFPE or frozen tumors of known and unknown histologies. Total RNA was analyzed, reverse-transcribed and used for the synthesis of labeled aRNA after two rounds of amplification. Labeled aRNA was hybridized to a 3'-based 22K spot oligonucleotide arrays, and compared to a labeled reference by two-color microarray analysis. After normalization, gene expression profiles were compared by unsupervised hierarchical clustering. Using this approach, at least 24% of unselected FFPE samples produced RNA of sufficient quality for microarray analysis. From our initial studies, we determined criteria based on spectrophotometric analyses and a novel TaqMan-based assay to predict which samples were of sufficient quality for microarray analysis before hybridization. These criteria were validated on an independent set of tumors with a 100% success rate (20 of 20). Unsupervised analysis of informative gene expression profiles distinguished tumor type and subtype, and identified tumor tissue of origin in three unclassified carcinomas. Although only a minority of FFPE blocks could be analyzed, we show that informative RNA expression analysis can be derived from selected FFPE samples.