Abstract: A novel Fourier descriptor based on multi-level triangular area functions (MTA) was proposed for shape retrieval. For each point on the shape contour, MTA values were derived from unequal-arc-length partitions of the shape contour. The MTA can finely capture the global features and local contour variations of the contour and their low-frequency Fourier coefficients were regarded as the feature vector for shape description. The image retrieval performance of the proposed method was evaluated on the standard MPEG-7 shape database and compared with those of Fourier descriptors derived from the centroid distance function, area function, farthest point distance function, angular radius function, arc-height radius complex function, and the combined Fourier descriptor. Experimental results demonstrate that the proposed method reaches the highest precision at the same recall value and has low complexity among these descriptors, showing its effectiveness.