Financial networks are constructed from asset price comovements. There is a large literature that takes these networks as given, for example, for portfolio optimization. But what exactly is the origin of these networks? We exploit a unique database with matched asset price and order imbalance data, allowing us to observe the trade orders placed and reveal excess demands along with the resulting prices. Empirically, we find that order imbalance comovement has a positive and statistically significant effect on return comovement. Filtering out the latent market factors from both order imbalance and return leads to a drastic drop in explanatory power. We infer that the market factor of order imbalance is the primary driver of return comovement—robust to model specifications as well as fixed effects. We present complementary results with market volatility and the decomposition of traders in terms of strategic heterogeneity. Our work brings forth the role of order imbalance networks in explaining asset return networks.