The Devaux-Vermeulen-Li (DVL) method is a simple method that directly retrieves aerosol optical parameters based on single-wavelength solar-sky radiation observations, without the assumption of aerosol microphysical properties. Inheriting the previous retrieval of single-scattering albedo (SSA) and scattering phase function, the DVL method is modified to derive the aerosol asymmetric factor (g) parameter. Interpolation methods were proposed to estimate the phase function over the extreme forward and backward scattering regions where instrumental observations are missing; thus, g could be derived from the phase function over the entire scattering angle region. To evaluate the g accuracy from the DVL algorithm, especially for non-spherical aerosols, synthetic retrieval with typical aerosol models (water-soluble, biomass burning, dust-sphericity, and dust-spheroid models) and retrievals from AERONET observations (Beijing site, from January 2011 to March 2015) were implemented at four wavelengths (440, 675, 870, and 1020 nm). The numerical experiments showed that DVL retrieves g with errors less than +/- 0.02 under "error-free" conditions. When measurement uncertainties were present, all g errors were within +/- 0.03, except for the angular pointing error for the coarse mode-dominated dust-sphericity/spheroid aerosols. Most importantly, g retrievals were not sensitive to the aerosol optical depth (AOD) and sky radiance errors, which are important influencing factors for SSA retrieval. Comparison of DVL retrievals with AERONET version 2 level 2.0 products [with AOD (440 nm) > 0.2] shows that the DVL g was well correlated with that of AERONET, especially for the 675, 870, and 1020 nm bands, with root mean square deviations (RMSDs) smaller than 0.02 and absolute values of mean bias deviation smaller than 0.01. Relatively larger deviations occurred at the 440 nm band, where g values were underestimated by approximately 0.03 compared to those of AERONET, with a higher RMSD of approximately 0.035. Both the synthetic retrieval and comparison with AERONET indicated that the algorithm performance for large, non-spherical particles is comparable to that of other spherical aerosol particles in retrieving g. (C) 2017 Optical Society of America