In this paper we present a system towards the recognition of off-line Oriya handwritten characters. Since most of the Oriya characters have curve-like stroke, we use curvature feature for the recognition purpose. To get the feature, at first, the input image is size normalized and segmented into 49?49 blocks. Curvature is then computed using bi-quadratic interpolation method and quantized into 3 levels according to concave, linear and convex regions. Next direction of gradient is quantized into 32 levels with /16 intervals, and strength of the gradient is accumulated in each of the 32 directions and in each of the 3 curvature levels of every block. A spatial resolution is made to get 7?7 blocks from 49?49 blocks and a directional resolution is made to get 8 directions from 32 directions. Using curvature features for 3 levels we get 1176 (7?7 blocks ? 8 directions ? 3 levels) dimensional features. Finally using principal component analysis we reduce the dimension 1176 to 392 and this 392 dimensional feature vector is fed to a quadratic classifier for recognition. We tested 18190 samples of Oriya handwritten samples and obtained 94.60% accuracy from our proposed system.