The sensitivity of nonlinear ultrasonic features to small crack is recognized much higher than what can be achieved by conventional linear features. However, in most of cases, the conventional nonlinear ultrasonic features can only locate the crack but difficult to quantify. In this paper, a novel technique for the quantification detection of small crack is presented. The structure needing to be detected is radiated by laser to produce a nonlinear ultrasound and its state space predictive model is constructed based on its dynamical responses in time domain. Then the Bhattacharyya Distance is adopted as the quantification metric to detect the crack. Furthermore, the crack is quantified using the feature extracted from the entire scanned area. An example of the approach and experimental measurement is given for alloys with different kinds of cracks. The experimental results show the reliability of the approach.