A cross-sectional study was undertaken to report prevalence and to identify risk factors of subclinical mastitis of dairy cattle in Sylhet district of Bangladesh. Among 325 dairy farms of the district 12 farms(3.7%) were selected conveniently for this study. All the dairy cows of the 12 farms were selected for sample collection. Fresh milk samples from each of the selected dairy cows were collected aseptically in separate sterilized test tube as RF, RH, LF and LH quarter of the udder. Rapid modified White Side Test(WST) was used to detect subclinical mastitis(SCM). Results of WST and data derived from filled in questionnaire were entered in Microsoft Excel 2003 and transferred to $STATA^{(R)}$, version 8.0/Intercooled(Stata Corporation, Texas, USA, 2003). The overall prevalence of SCM and its distribution in different categories of variables in cow and their exact binomial 95% confidence intervals were calculated in $STATA^{(R)}$. Simple bivariable associations among independent variables were investigated by $x^2$ test in $STATA^{(R)}$. Multiple logistic regression analysis with backward elimination method was used to identify risk factors of SCM. To identify significant variation in quarter SCM, linear regression analysis was performed after arcsine transformation of the data. The overall prevalence of SCM found in this study is 54%. Dairy cows with teat lesions had significantly increased SCM(OR=12342, P value=0.000, 95% CI=762, 199798) than others without teat lesions. The Holstein Friesian X Jersey X Sahiwal breed has significantly decreased(OR=0.18, p=0.03, 95% CI 0.04, 0.85) SCM than other breeds. The prevalence of SCM found in this study is in agreement with others. The injury in the teat increases the probability of getting infected with microbes and thereby mastitis. If the prevalence of teat lesion can be decreased the probability of subclinical mastitis will also be decreased. The negatively associated Holstein Friesian X Jersey X Sahiwall breed may help in planning mastitis control program if this finding can be validated by a more powerful case-control or cohort study design.