The North American West Nile virus (WNV) epizootic, which began in 1999, has caused significant morbidity and mortality in horses. Because experimental infection has failed to consistently produce encephalitis in inoculated horses, investigation of naturally occurring cases was used to optimize strategies for diagnosis of this disease. Although WNV RNA could be detected by reverse transcriptase-polymerase chain reaction (RT-PCR) performed on whole blood collected from both clinically affected horses and unaffected herdmates, the diagnostic sensitivity of this approach was low compared with IgM-capture enzyme-linked immunosorbent assay. In addition, it was observed that 18.5% of herdmates of clinically ill horses seroconverted to WNV yet exhibited no overt clinical signs of WNV encephalitis. West Nile viral RNA was detected in neural tissue of 46 of 64 dead horses that were suspected of having WNV encephalitis. Some of these animals were IgM negative or had not been tested serologically. A primary cause of death other than WNV encephalitis was identified in 15 of the 64 cases, whereas the final diagnosis for 3 of these cases remains unresolved. Quantitative RT-PCR analysis of neural tissue from WNV RNA-positive horses demonstrated that the medulla contained the highest mean concentration of viral RNA and that WNV RNA could be detected in samples extracted from formalin-fixed neural tissue. A comparison of WNV RT-PCR amplification strategies found that nested RT-PCR improved diagnostic sensitivity only slightly over a single round of amplification and that a quantitative (TaqMan) assay had sensitivity and specificity that were equivalent to those of nested amplification.