Abstract: In order to solve the problems that the analytical precision met the standard requirements when detecting sulfur in steel according to the standard methods GB/T 4336-2016 and GB/T 14203-2016 while it would decrease significantly with the increase of manganese content in steel, scanning electron microscopy (SEM) was used for exploring reasons, and optimization of the analysis channel, excitation parameters and iron reference line of spark source direct reading spectrometer was used for improving questions. After the sample surface was cleaned and milled, the standard method GB/T 4336-2016 was applied to 5 replicate analysis of standard samples (standard sample 1^# with manganese mass fraction of 2.00% and standard sample 2^# with manganese mass fraction of 0.043%). RSD (2.5%) of the determined values of sulfur element in standard sample 1^# was significantly greater than that (1.2%) in standard sample 2^#, which might be related to the fact that the sulfide in standard sample 1^# was manganese sulfide and manganese sulfide was unevenly distributed in the inclusions. The following parameters were adjusted to improve the above problems:the prespark time was 6 s, the exposure integration time was 1 s, the analysis channel was S1, and the reference line of iron element was 136.137 nm. It was shown that the linear range of the working curve of sulfur was found in the range of 0.001 0%-0.117%, with detection limit of 0.000 06%. This method was applied to analysis of the sample with manganese mass fraction of 2%, RSD (n=6) of the determined values was less than 1.0%, and the determined values were consistent with those obtained by the carbon sulfur analyzer.