Vibration of the heart wall arises from motion due to myocardial contraction and relaxation, and thus this vibration reflects the function of the heart. The activity of the heart muscle can be estimated by investigating the spectrum transition of spatial and/or frequency spectra between vibration signals. It is difficult, however, to estimate the spectrum transition patterns of nonstationary signals, such as vibration, and to diagnose heart diseases. A new method of the estimation of spectrum transition is presented by introducing a constraint into the cost function, which is described by the partial correlation coefficients, by constraint least-squares method. By applying this method to the analysis of vibration signals, noninvasively measured by a method developed in our laboratory using ultrasonics, on the left ventricular side of the interventricular septum, the spectrum transition during a period of one beat was investigated. Furthermore, by analysis of some vibration signals on the vertical cross section of the interventricular spectrum between the right and left ventricles, the spectrum transition of the myocardial layers was investigated and the spectrum transition patterns of normal individuals and patients were clearly obtained. The establishment of this method opens up a new way of diagnosing heart diseases.