The philosophy of operation of any algorithm based on decision trees is quite simple. In fact, although sometimes containing important differences in the way to do this or that step, any algorithm of this category is based on the strategy of divide and conquer. In general, this philosophy is based on the successive division of the problem into several subproblems with a smaller number of dimensions, until a solution for each of the simpler problems can be found. Based on this principle, the classifiers based on decision trees try to find ways to divide the universe into successively more subgroups (creating nodes containing the respective tests) until each addressing only one class or until one of the classes shows a clear majority do not justifying further divisions, generating in this situation a leaf containing the class majority....
more details here,
http://en.wikibooks.org/wiki/Data_Mining_Algorithms_In_R/Classification/Decision_Trees
http://ocw.mit.edu/courses/sloan-school-of-management/15-097-prediction-machine-learning-and-statistics-spring-2012/lecture-notes/MIT15_097S12_lec08.pdf
