The paper presents an analysis of time series concerning diurnal differences between standard and automatic results of measurements of three air thermal parameters – mean (Tm), maximum (Tmax) and minimum (Tmin) temperature. By means of artificial neural networks an attempt at identification of the occurring patterns and their repetition was made. The study was conducted in the period of 2000-2009 on the site of Agro-and Hydrometeorology Observatory Wroclaw-Swojec which belongs to the Wroclaw University of Environmental and Life Sciences. In order to achieve the intended aim, single-layer perceptron networks were used, created in MATLAB (Neural Network Toolbox) and STATISTICA 10. The following aspects were subjected to the analysis: the architecture of the developed networks, the number of cycles in the learning process, the changes in Mean Squared Error (MSE), and the correlations between the values of the parameters obtained by means of meteorological instruments and the ones prognosticated by the networks. Despite multiple attempts, no model of satisfying quality was obtained. It was concluded that the nature of the time series was that of white noise, meaning that the occurrence of differences between both air temperature measurement methods were characterised by randomness and a lack of visible circularity and trend.