Sea surface wind field was retrieved from high-resolution SIR-C SAR data by using CMOD algorithms off the east coast of Korea. In order to extract wind direction information from SAR data, a two-dimensional spectral analysis method was applied to the normalized radar cross section of the image. An $180^{circ}$-ambiguity problem in the determination of wind direction was solved by selecting a direction nearest to the wind vector of the ECMWF reanalysis data. Comparison of the wind retrieval patterns with the ECMWF and NCEP/NCAR dataset showed RMS errors in the range of 1.30 to $1.72;ms^{-1}$. In contrast, comparison of wind directions revealed large errors of greater than $60^{circ}$, which is enormously higher than the permitted limit of about $20^{circ}$ for satellite scatterometer winds. Compared with wind speed results from different algorithms, wind vectors based on commonly-used CMOD4 algorithm showed good agreement with those derived by other algorithms such as CMOD_IFR2 and CMOD5, particularly at medium winds from 4 to $8;ms^{-1}$. However, apparent discrepancy appeared at low winds (< $4;ms^{-1}$). This study also addressed an importance of accurate wind direction data to improve the accuracy of wind speed retrieval and discussed potential causes of wind retrieval errors from SAR data.