GPU’s have recently emerged as a significantly more powerful computing plat-form, capable of several orders of magnitude faster computations compared toCPU based approaches. However, they require significant changes in the algorithmic design compared to traditional programming paradigms. In this chapter we specifically introduce the reader to an overview of GPGPU development tools and the potential algorithmic pitfalls and bottlenecks when developing medical imaging algorithms for the GPU. We present a few general methodologies and building blocks for implementing fast image processing on GPUs. More specifically they include: methods for performing fast image convolutions and filtering;line detection, and bandwidth and memory considerations when processing volumetric datasets. Finally we conclude with a discourse on numerical precision as well as on mixing single floating-point versus double floating-point code.