Analytic network process ( ANP) is considered, in this paper, to have three shortcomings. Firstly, the subjective judgment mode adopted in ANP, taking the logic of comparing object A with object B with respect to object A，does result in serious arbitrariness of decision makers in the comparisons of inner self-dependence relations within a cluster and in those of relative importance among clusters. Secondly, the weighting approach given in ANP to clusters is also arbitrary from the theoretical viewpoint. Thirdly, within ANP, no consideration of the structural difference among source cluster-elements, transient cluster-elements and sink cluster-elements is taken into account in the procedure of final ranking weighting computation. To solve these problems, a new analytic network called the cone network is presented, and a new theoretical analytic approach i.e., Cone-ANP, which is distinguishingly different from ANP，is further innovatively built through reconstructing the ANP theory and by carrying the corresponding mathematical deduction. Illustrative examples show that evaluation conclusions can be more reasonably drawn with Cone-ANP than with ANP.