We present a weak lensing mass map covering similar to 124 deg(2) of the Canada-France-Hawaii Telescope Stripe 82 Survey (CS82). We study the statistics of rare peaks in the map, including peak abundance, the peak-peak correlation functions and the tangential-shear profiles around peaks. We find that the abundance of peaks detected in CS82 is consistent with predictions from a I > cold dark matter cosmological model, once noise effects are properly included. The correlation functions of peaks with different signal-to-noise ratio (SNR) are well described by power laws, and there is a clear cross-correlation between the Sloan Digital Sky Survey III/Constant Mass galaxies and high SNR peaks. The tangential-shear profiles around peaks increase with peak SNR. We fit analytical models to the tangential-shear profiles, including a projected singular isothermal sphere (SIS) model and a projected Navarro, Frenk & White (NFW) model, plus a two-halo term. For the high SNR peaks, the SIS model is rejected at similar to 3 sigma. The NFW model plus a two-halo term gives more acceptable fits to the data. Some peaks match the positions of optically detected clusters, while others are relatively dark. Comparing dark and matched peaks, we find a difference in lensing signal of a factor of 2, suggesting that about half of the dark peaks are false detections.