This project investigates the manufacturing flow variability in order to stabilize the factory process flow. Nowadays, in manufacturing production lines and particularly in modern front end semiconductor lines, processes and equipments are very complex. Any disturbance of the process creates variability in the line, and causes substantial losses in productivity for manufacturing corporations. These disturbances are unpredictable, difficult to control and result in long recovery times. Variability occurring in a production system disturbs the whole processing flow and results in long product cycle times. Hence, a range of sources of variability was determined from the literature and analyzed. This lead with the cooperation of factory managers to the development of four main objectives: (1) Determine a proper metric to measure the variability in the production system. (2) Determine the effect of batching and tool availability on the process flow. (3) Understand the interaction between operations. (4) Develop a release strategy in order to stabilize the production flow. First, from the observation of real production data, a difference metric was developed and operations creating or removing variability were identified. The propagation of variability can be followed using a correlation coefficient. Nevertheless, the data were not detailed enough to explain the origin of the variability. Consequently, several simulation models were created to investigate variability. The simulations‘ results show that the release strategy should be adjusted as a function of batch, tool availability and constraint parameters, in order to stabilize the flow of items in the line and control cycle time and cycle time variability. The notion of critical availability is introduced and defined. Improvement of the line performance is obtained through a tighter control of the availability of high capacity operations. This lead to the development of a new hybrid push pull release strategy, named CONFLOW, to regulate the flow of items reaching the constraint operation. CONFLOW was tested under many simulating conditions (batching, parallel processing, and different line length). Compared to a push system, CONFLOW release strategy results, into significant improvement (up to 80%) in cycle time, cycle time standard deviation and WIP level at the cost of 13% reduction in throughput. CONFLOW performances were compared to common TOC strategies (SA and DBR). The results are encouraging. In the specific conditions considered, CONFLOW performances are similar to SA and slightly better than DBR.