Clinical diagnosis of prostate cancer is most often done by transrectal ultrasound-guided needle biopsy. Because of the low resolution of ultrasound, however, the urologist cannot usually distinguish between cancerous and healthy tissue. Therefore, most biopsies follow standard protocols based on long-term physician experience. Recent studies indicate that these protocols may have a significant rate of false negative diagnoses. This research develops optimized biopsy protocols. We use real prostate specimens removed by prostatectomy to develop a 3D distribution map of cancer in the prostate. We develop also a probability model of the needle insertion procedure. Using this model, the tumor map, and the geometry of the biopsy needle, we obtain estimates for the probability of obtaining a positive biopsy in various zones of prostates with cancer. We develop a nonlinear optimization problem that determines the protocols that maximize the probability of cancer detection for a given number of needles, and present new optimized protocols.