Background: Neutrino-driven winds following core-collapse supernova explosions have been proposed as a possible site where light r-process nuclei (between Fe and Ag) might be synthesized. In these events, (α, n) reactions are key to move matter towards the region of higher proton number. Abundance network calculations are very sensitive to the rates for this type of reactions.Purpose: The present work aims at evaluating the theoretical uncertainty of these (α, n) reactions calculated with reaction codes based on the Hauser-Feshbach model.Method: We compared several (α,n) rates taken from TALYS and the NON-SMOKER database to determine the uncertainties due to the existing technical differences between both codes. In addition, we evaluated the sensitivity of TALYS rates to variations in the alpha optical potentials, masses, level densities, optical potentials, pre-equilibrium intranuclear transition rates, level structure, radiative transmission coefficients, and width-fluctuation correction factors.Results: The main source of uncertainty at low temperatures are mostly due to the use of different alpha optical potentials. Differences between TALYS and NON-SMOKER at high temperatures arise from the energy binning algorithm used by each code. We have also noticed that the (α,n) rates from the NON-SMOKER database correspond to the inclusive reaction, instead of the exclusive (α,1n) channel calculated in the present work and used in network calculations.Conclusions: Theoretical uncertainties in calculated reaction rates can be as high as 1-to-2 orders of magnitude, and strongly dependent on the temperature of the environment. Besides direct measurements of the inclusive and exclusive (α, 1n) reaction rates, experimental studies of alpha optical potentials are crucial to improve the performance of reaction codes.