We present an algorithm implemented in the astroalign Python module for image registration in astronomy. Our module does not rely on WCS information and instead matches 3-point asterisms (triangles) on the images to find the most accurate linear transformation between the two. It is especially useful in the context of aligning images prior to stacking or performing difference image analysis. Astroalign can match images of different point-spread functions, seeing, and atmospheric conditions.