A substantial share of the national budget in developing countries goes towards subsidies for the needy. The two most common methods to identify beneficiaries are a rule-based allocation relying on a proxy-means test (PMT) or a decentralized allocation in which a local leader selects beneficiaries. A decentralized allocation may offer informational or accountability advantages, but may be prone to elite capture. We study this tradeoff in the context of two large-scale subsidy programs (one for agricultural inputs and one for food) in Malawi, both of which were decentralized to traditional leaders or “chiefs”. We find that chiefs are more likely to channel subsidies to kin, but that chiefs target needy households (as proxied by consumption) nearly as well as (or, not much worse than) even a perfectly respected PMT. What's more, results from a model-based test are consistent with chiefs taking into consideration productive efficiency when identifying beneficiaries for input subsidies. Thus some of the poverty-mistargeting of input subsidies could be welfare improving, since income pooling at the village level appears common.