3D neuron terminal points could be very good seed points in neuron tracing algorithms. Previously, a rayshooting model to detect neuron terminal point was proposed by analyzing the intensity distribution characteristics of the neighborhoods around the terminal point candidates. However, the length of the shooting rays and the number of zslices that should be considered in this model are fixed, its accuracy would be seriously affected when handling datasets where the diameter of the neuron varies much. Thus, an adaptive rayshooting model is proposed by changing the length of the shooting rays and the number of adjacent slices according to the local diameter of the neuron obtained by the MSFM (multistencils fast marching) method and Rayburst sampling algorithm. Compared with the previous work, the experimental results show that the proposed method could improve the detection accuracy by about 10%.