A method for determining articulatory parameters from speech acoustics is presented. The method is based on a search of an articulatory-acoustic codebook which is designed from simultaneous observation data of articulatory motions and speech acoustics. The codebook search employs dynamic constraints on acoustic behavior as well as articulatory behavior. There are two constrains. One of the constraints is use of spectral segments in the codebook search and the other is use of the smoothness of articulatory trajectories in the articulatory parameter path search. The articulatory parameters are determined by selecting the articulatory code vector in the codebook which minimizes the weighted distance measure of segmental spectral distance and squared distance between succeeding articulatory parameters. The results of an experiment show that an rms error between the estimated and observed articulatory parameter was about 2.0 mm on average, and the articulatory features for vowels and consonants are recovered well.