This paper presents a methodology of creating a statistical atlas of spatial distribution of prostate cancer from a large patient cohort, and uses it for designing optimal needle biopsy strategies. In order to remove inter-individual morphological variability and determine the true variability in cancer position, an adaptive-focus deformable model (AFDM) is used to register and normalize prostate samples. Moreover, a probabilistic method is developed for designing optimal biopsy strategies that determine the locations and the number of needles by optimizing cancer detection probability. Various experiments demonstrate the performance of AFDM in registering prostate samples for construction of the statistical atlas, and also validate the predictive power of our atlas-based optimal biopsy strategies in detecting prostate cancer.