Conventional fluid flow estimation methods for in vivo optical microscopy are limited to two-dimensions and are only able to estimate the components of flow parallel to the imaging plane. This limits the study of flow in more intricate biological structures, such as the embryonic zebrafish heart, where flow is three-dimensional. To measure three-dimensional blood flow, we propose an algorithm to reconstruct a 3D, divergence-free flow map from multiple 2D flow estimates computed from image stacks captured from different views. This allows us to estimate the out-of-plane velocity component that is normally lost with single-view imaging. This paper describes our 3D flow reconstruction algorithm, evaluates its performance on a simulated velocity field, and demonstrates its application to in vivo cardiac imaging within a live zebrafish larva.