An improved method of correcting for the angle of attack error resulting from the imperfect (co)sine response of ultrasonic anemometers is proposed. The angle of attack, which was calculated as the arctangent of observed wind vectors, contains the angle of attack errors in the vectors themselves, and hence this angle was 'false'. The 'true' angle of attack should be calculated from the corrected or 'true' wind vectors. In the improved method, the 'true' angle of attack is derived by solving a nonlinear equation which connects the 'false' angle of attack to the 'true' one. In applying this method to the case of R2- and R3-type Solent ultrasonic anemometers, the fit of the function for the sine responses to wind tunnel data was improved, and the cosine response function was also improved to consider the effect of the difference of the vertical positions of the transducers. The nonlinear equation was solved using the Steffensen method; robustly and adequately fast for practical use in calculating eddy fluxes. The accuracy of the correction method is improved over a previous one, especially at large angles of attack. Applying our correction method to field data from two forests and one peat bog, the eddy fluxes of sensible heat, latent heat and CO