Abstract: In this paper, we present a GPU-based algorithm for reconstructing 3D wavelets using fragment programs. To minimize the data transfer and fragment processing overhead, we propose a novel scheme that uses tileboards as a primary layout to organize 3D wavelet coefficients. By accessing the tileboards with correct texture coordinates, Haar and Daubechies wavelets can be evaluated by the GPU in real time. The tileboard also serves as input to the rendering programs. We demonstrate how the tileboards allow us to efficiently cull unnecessary data, and we extend our work to render large volumes with multiple resolution levels.