Background: Active surveillance (AS) allows patients with localized prostate cancer to delay and even avoid treatments that carry risk of significant side-effects. However, biomarkers are needed to more accurately risk-stratify patients and enhance acceptance of AS. Challenges with biomarkers measured from needle biopsy specimen that need to be address include limited material, tumor and specimen heterogeneity and fragmentation of RNA. Methods: A total of 23 patients with matched formalin-fixed paraffin-embedded (FFPE) biopsy and radical prostatectomy (RP) specimens were identified at University of California, San Francisco and Cedar Sinai. For each patient, samples were taken from tumor, non-neoplastic tissue adjacent to tumor (NAT) and non-neoplastic tissue contralateral to tumor (NCT) from both biopsy and RP specimens. RNA expression was measured in 130 samples using the 1.4 million feature Affymetrix Human Exon 1.0 ST arrays. Results: From 1 mm cylindrical cores punched from FFPE blocks, sufficient RNA was extracted for the assay from 63 out of 69 (91%) biopsy specimens and 69 out of 69 (100%) RP specimens. Median RNA yield from biopsies was lower than RP specimens but a similar quantity and quality of cDNA was amplified from the 100 ng RNA required for the assay. RNA from 62/63 (98%) of the biopsy specimens and 68/69 (99%) of the RP specimens generated expression data that passed quality control. Genomic classifiers (GC) predicting metastasis (GC1) and presence of high Gleason grade tumor (GC2) were highly correlated between matched biopsy and RP specimens with R=0.74 (p=0.0003) and R=0.63 (p=0.004), respectively. For GC2, matched tumor and NAT from RP had a higher correlation (R=0.57, p=0.0045) than matched tumor and NCT (R=0.10, p=0.67). There was no relationship between percentage of stromal contamination in the biopsy and the GC scores for either classifier, R=0.02 (p=0.91) and R=-0.16 (p=0.28) respectively. Conclusions: RNA expression levels can be measured in FFPE needle biopsy on a genome-wide scale with similar data quality compared to RP specimens. GC scores from biopsies and RP samples correlate well, are robust to stromal contamination, and are present in adjacent non-neoplastic tissue.