Terrestrial laser scanning is a new technology uprising in 1990's. Owing to its ability to obtain a large number of 3D coordinates of points, called point clouds, in real time, it has been attracted attentions of surveying field. To build threedimensional models, multiple scans from different viewpoints are required due to occluded surfaces and limited field of view of the scanner. These multi-view point clouds then must be transformed into a common reference frame in order to describe complete object being researched. This process is called registration of point clouds. Pairwise registration often is a common method for registration of multi-views point clouds. Owing to characteristics of error accumulation and propagation, using pairwise registration method will result in severely distortion of researched object. In order to overcome this shortcoming, we present a new registration method. Firstly, we use pairwise registration method to calculate transformation parameters of adjacent two scans, and by selecting coordinate system of first scan as a uniform coordinate system, we connect all point clouds into a loosely network. Secondly, according to condition that common points between first scan and last scan must have the same coordinates in uniform system, we use nonlinear correction model to compute the distortion parameters of loosely network and lastly we correct distortion of each single point clouds and determine the best position of each point. Experiment is carried out and the results show that the registration error has reduced from 1.7cm to 5mm after correction, which demonstrates correctness of the method.