This paper proposes a concept of non-intrusive load monitoring system for smart meter to monitor the situation of loads. In this study, the user can clearly know the power consumption of loads by observing the operation and time of use of loads, and then improve the habit of consumption to complete the goals of saving energy and reducing carbon. This paper employs a scheme of non-intrusive load monitoring system by extracting the significant and representative power signatures of voltage and current at utility service entry in identifying loads and analyzing the characteristics of loads, and then finds out the physical behavior of operation of loads to establish the model of loads. This paper uses short-time Fourier transform (STFT) and wavelet transform (WT) of time-frequency domain to analyze and compare different loads in the experiments. In the experiments, the results reveal wavelet transform is better than STFT on transient analysis of loads. Choice of power signatures affects the results of load recognition and computation time.