The triple oxygen isotopic composition of dissolved oxygen (17?) is a promising tracer of gross oxygen productivity (P) in the ocean. Recent studies have inferred a high and variable ratio of P to 14C net primary productivity (12–24 h incubations) (e.g., P:NPP(14C) of 5–10) using the 17? tracer method, which implies a very low efficiency of phytoplankton growth rates relative to gross photosynthetic rates. We added oxygen isotopes to a one-dimensional mixed layer model to assess the role of physical dynamics in potentially biasing estimates of P using the 17? tracer method at the Bermuda Atlantic Time-series Study (BATS) and Hawaii Ocean Time-series (HOT). Model results were compared to multiyear observations at each site. Entrainment of high 17? thermocline water into the mixed layer was the largest source of error in estimating P from mixed layer 17?. At both BATS and HOT, entrainment bias was significant throughout the year and resulted in an annually averaged overestimate of mixed layer P of 60 to 80%. When the entrainment bias is corrected for, P calculated from observed 17? and 14C productivity incubations results in a gross:net productivity ratio of 2.6 (+0.9 ?0.8) at BATS. At HOT a gross:net ratio decreasing linearly from 3.0 (+1.0 ?0.8) at the surface to 1.4 (+0.6 ?0.6) at depth best reproduced observations. In the seasonal thermocline at BATS, however, a significantly higher gross:net ratio or large lateral fluxes of 17? must be invoked to explain 17? field observations.