This report describes how blob features can be used for automatic estimation of the fundamental matrix from two perspective projections of a 3D scene. Blobs are perceptually salient, homogeneous, compact image regions. They are represented by their average colour, area, centre of gravity and inertia matrix. Coarse blob correspondences are found by voting using colour and local similarity transform matching on blob pairs. We then do RANSAC sampling of the coarse correspondences, and weight each estimate according to how well the approximating conics and colours of two blobs correspond. The initial voting significantly reduces the number of RANSAC samples required, and the extra information besides position, allows us to reject false matches more accurately than in RANSAC using point features.