In this study considering total suspended solids (TSS) parameter monitored in a stream watershed, the predictability of upstream values from downstream data was investigated using regression analysis, which were applied to linear, power, exponential, and quadratic functions, and artificial neural networks (ANNs) method. The data were obtained within the scope of sampling studies carried out 40 times between June 2019 and March 2020 at eight monitoring stations selected in the Sera Stream Watershed (Trabzon). The monitoring stations were divided into two groups as upstream, the first four, and downstream, the last four, stations. Half of downstream data (two stations) was used for training, a quarter (one station) for validation, and the rest (one station) for testing. Two models with different combinations of independent variables were established. In the first model (M1), only the TSS values, and in the other model (M2), the month and week information of the sampling dates were digitized and used as independent variables, in addition to the TSS values. Root mean square error, mean absolute error, and Nash-Sutcliffe (NS) efficiency coefficient statistics were used to evaluate the model and method performances. Compared to other functions, the power one had the best estimation results in the regression analysis. On the other hand, the ANNs method gave better results than the regression analysis. In both methods, M2 performed better overall. In the ANNs method, the NS efficiency coefficients obtained from M1 and M2 were calculated as 0.980 and 0.997, respectively, for the training, and 0.978 and 0.978, respectively, for the testing data sets. Considering the efficiency values, it has been understood that the use of date information as an independent variable will positively affect the model performance in the stream TSS modeling studies. Within the scope of this study, it has been concluded that upstream TSS values can be successfully estimated from downstream TSS data in stream watersheds.