Gene expression profiling (GEP) has identified genes whose expression levels predict patient survival in diffuse large-B-cell lymphoma (DLBCL). Such discovery techniques generally require frozen samples unavailable for most patients. We developed a quantitative nuclease protection assay to measure expression levels of prognostic DLBCL genes using formalin-fixed, paraffin-embedded (FFPE) tissue. FFPE tissue was sectioned, permeabilized, denatured in the presence of specific probes, and hybridized to mRNA in situ. Nuclease subsequently destroyed non-hybridized probe. Alkaline hydrolysis freed mRNA-bound probes from tissue, which were transferred to ArrayPlates for probe capture and chemiluminescent quantification. We validated assay performance using frozen, fresh, and FFPE DLBCL samples, then used 39 archived DLBCL, previously microarray analyzed, to correlate GEP and ArrayPlate results. We compared old (>18 years) with new (<2 months) paraffin blocks made from previously frozen tissue from the original biopsy. ArrayPlate gene expression results were confirmed with immunohistochemistry for BCL2, BCL6, and HLA-DR, showing agreement between mRNA species and the proteins they encode. Assay performance was linear to approximately 1 mg sample/well. RNase and DNase treatments demonstrated assay specificity for RNA detection, both fixed and soluble RNA detection. Comparisons were excellent for lysate vs snap-frozen vs FFPE (R(2)>0.98 for all comparisons). Coefficients of variation for quadruplicates on FFPE were generally <20%. Correlation between new and old paraffin blocks from the same biopsy was good (R(2)=0.71). Comparison of ArrayPlate to Affymetrix and cDNA microarrays showed reasonable correlations. Insufficient power from small sample size prevented successfully correlating results with patient survival, although hazard ratios trended the expected directions. We developed an assay to quantify expression levels of survival prediction genes in DLBCL using FFPE, fresh, or frozen tissue. While this technique cannot replace GEP for discovery, it indicates that expression differences identified by GEP can be replicated on a platform applicable to archived FFPE samples.