This paper presents the application of artificial neural networks (ANNs) and regression analysis (RA) for predicting dissolved oxygen concentrations (DO, mg/L) from water quality (WQ) indicators, namely stream water pH and temperature (t, ºC). For this purpose, three diverse models are used in our analysis, considering the functional relationship between in situ-measured WQ indicators and DO concentration. The WQ data are semimonthly obtained from nine monitoring sites in the Harsit Stream watershed in the Eastern Black Sea Basin of Turkey, from March 2009 to February 2010. As a result of model prediction, this study proposes a suitable ANN model, including two independent variables to efficiently predict DO concentration from WQ data, with the root mean square error of 0.9442 mg/L and mean absolute error of 0.6965 mg/L. The proposed model predicts the DO concentration better than the RA and the other two ANN models. The results may reduce the time and cost necessary to determine DO concentrations.