Recently, the ability to predict spike times of biological neurons has become an important evaluation criterion of spiking neuron models[1]. In the process of spike time prediction, we optimize a model by using only a part of experimental data obtained from real neurons, and we quantitatively evaluate the model by a proportion of the number of real spikes that the model succeeded to reproduce their times in the rest of the data. In this study, we focus on Multi-timescale Adaptive Threshold model (MAT) proposed by Kobayashi et al[2], which has a great ability to predict the spike timing. We analyzed the dynamical behavior of MAT in response to simple current pulses, and qualitatively evaluate its ability to reproduce firing patterns of biological neurons. We revealed that MAT can reproduce a variety of firing patterns despite that MAT consists of a few simple linear formulas and is a hard thresholding model similar to leaky integrate-and-fire model. These results, we expect, could provide us to discuss about some relationship between the qualitative ability to reproduce firing patterns[3] and the quantitative ability to predict spike times, and give a clue to a new comprehensive way of the assessment and the classification of the single neuron models.[1] Gerstner W., Naud R., Science 326: 379-380 (2009)[2] Kobayashi R., Tsubo Y., Shinomoto S., Front. Comput. Neurosci. 3:9 (2009) [3] Izhikevich, E. M. IEEE Trans. Neural Netw. 15, 1063–1070 (2004)