ART is an iterative algebraic process to solve linear equations. Its basic idea is to back-project the error projections proportional to the estimated pixels. OS-EM is a very efficient method to maximize the likelihood criterion, using ordered subsets technique. In our generalized iterative reconstruction techniques (GIRT) two procedures are suggested, one is to choose a method of error back-projection and the other is to choose a way to average the pixel's increment within every specified subset. From this point of view, ART, ML-EM, OS-EM and RAMLA are special cases of GIRT. In addition, a new image reconstruction algorithm with very good convergent property has been obtained by using a special error back-projection method-Proportional to the pixel's contribution. This new algorithm can also be combined with ordered subsets method and Gibbs smoothing, which can make the iterative process much faster.