Diffusion is a well-known algorithm for load-balancing in which tasks move from heavily-loaded processors to lightly- loaded neighbors. This paper presents a rigorous analysis of the performance of the diffusion algorithm on arbitrary networks. It is shown that the running time of the diffusion algo- rithm is bounded by: and (log) ≤ Time <0(N) < n(log) ≤ Time ≤ 0(2), M where N is the number of nodes in the network, σ is the standard deviation of the initial load distribution (which represents how imbalanced the load is initially), and I and are the network's electrical and fluid conductances respec- tively (which are measures of the network's bandwidth). For the case of the generalized mesh with wrap-around (which includes common networks like the ring, 2D-torus, 3D-torus and hypercube), we derive tighter bounds and con- clude that the diffusion algorithm is inefficient for lower di- mensional meshes.