This paper presents a new method to estimate HMM parameters of an acoustical transfer function based on HMM decomposition in model domain. The model parameters are estimated by maximizing a likelihood of adaptation data. The proposed method is obtained as the natural result of a reverse process of the HMM composition. In our previous paper[l], we proposed a method which can model an observed signal by the composition of HMMs of clean speech, noise and an acoustical transfer function. The previously proposed method needs measurement of impulse responses. It is inconvenient and unrealistic to measure impulse responses for a new environment. The new method is able to estimate HMM parameters of the acoustical transfer function from a small amount of adaptation data. Its effectiveness is confirmed by a series of speaker dependent and independent word recognition experiments on simulated distant-talking speech.