We use a global three-dimensional model simulation of atmospheric methanol to examine the consistency between observed atmospheric concentrations and current understanding of sources and sinks. Global sources in the model include 128 Tg yr-1 from plant growth, 38 Tg yr-1 from atmospheric reactions of CH3O2 with itself and other organic peroxy radicals, 23 Tg yr-1 from plant decay, 13 Tg yr-1 from biomass burning and biofuels, and 4 Tg yr-1 from vehicles and industry. The plant growth source is a factor of 3 higher for young than from mature leaves. The atmospheric lifetime of methanol in the model is 7 days; gas-phase oxidation by OH accounts for 63% of the global sink, dry deposition to land 26%, wet deposition 6%, uptake by the ocean 5%, and aqueous-phase oxidation in clouds less than 1%. The resulting simulation of atmospheric concentrations is generally unbiased in the Northern Hemisphere and reproduces the observed correlations of methanol with acetone, HCN, and CO in Asian outflow. Accounting for decreasing emission from leaves as they age is necessary to reproduce the observed seasonal variation of methanol concentrations at northern midlatitudes. The main model discrepancy is over the South Pacific, where simulated concentrations are a factor of 2 too low. Atmospheric production from the CH3O2 self-reaction is the dominant model source in this region. A factor of 2 increase in this source (to 50--100 Tg yr-1) would largely correct the discrepancy and appears consistent with independent constraints on CH3O2 concentrations. Our resulting best estimate of the global source of methanol is 240 Tg yr-1. More observations of methanol concentrations and fluxes are needed over tropical continents. Better knowledge is needed of CH3O2 concentrations in the remote troposphere and of the underlying organic chemistry.