The concept of pull production systems, such as kanban and CONWIP (Constant-Work-In-Progress), has been proposed for a long time. Although they have been implemented at some production systems with positive outcomes, some failure cases have also been reported. To find out the reason, a fundamental study of pull systems is conducted through queueing theory in this work. The theory of pull systems is developed from the perspective of information. The main advantage of kanban is to facilitate synchronization in an assembly line. However, consistent with prior observations, we find productivity improvement is achieved through variability reduction rather than kanban, and both kanban and CONWIP will reduce system capacity. To enjoy the benefit of kanban and avoid its shortcomings, a proactive pull system based on queue times is proposed. Numerical examples are given to illustrate the concepts. The implementation and benefit of proactive pull systems are explained through a case study in a semiconductor fab.