Background: Classical novae are cataclysmic nuclear explosions occurring when a white dwarf in a binary system accretes hydrogen-rich material from its companion star. Novae are partially responsible for the galactic synthesis of a variety of nuclides up to the calcium (A ∼ 40) region of the nuclear chart. Although the structure and dynamics of novae are thought to be relatively well understood, the predicted abundances of elements near the nucleosynthesis endpoint, in particular Ar and Ca, appear to sometimes be in disagreement with astronomical observations of the spectra of nova ejecta. Purpose: One possible source of the discrepancies between model predictions and astronomical observations is nuclear reaction data. Most reaction rates near the nova endpoint are estimated only from statistical model calculations, which carry large uncertainties. For certain key reactions, these rate uncertainties translate into large uncertainties in nucleosynthesis predictions. In particular, the 38 K(p,γ ) 39 Ca reaction has been identified as having a significant influence on Ar, K, and Ca production. In order to constrain the rate of this reaction, we have performed a direct measurement of the strengths of three candidate = 0 resonances within the Gamow window for nova burning, at 386 ± 10 keV, 515 ± 10 keV, and 689 ± 10 keV. Method: The experiment was performed in inverse kinematics using a beam of unstable 38 K impinged on a windowless hydrogen gas target. The 39 Ca recoils and prompt γ rays from 38 K(p,γ ) 39 Ca reactions were detected in coincidence using a recoil mass separator and a bismuth-germanate scintillator array, respectively. Results: For the 689 keV resonance, we observed a clear recoil-γ coincidence signal and extracted resonance strength and energy values of 120 +50 −30 (stat.) +20 −60 (sys.) meV and 679 +2 −1 (stat.)±1(sys.) keV, respectively. We also performed a singles analysis of the recoil data alone, extracting a resonance strength of 120 ± 20(stat.)±15(sys.) meV, consistent with the coincidence result. For the 386 keV and 515 keV resonances, we extract 90% confidence level upper limits of 2.54 meV and 18.4 meV, respectively. Conclusions: We have established a new recommended 38 K(p,γ ) 39 Ca rate based on experimental information, which reduces overall uncertainties near the peak temperatures of nova burning by a factor of ∼250. Using the rate obtained in this work in model calculations of the hottest oxygen-neon novae reduces overall uncertainties on Ar, K, and Ca synthesis to factors of 15 or less in all cases.