Present data stream management systems allow the automatic recording and processing of huge data volumes to guide any kind of process control or business decision. However, a crucial problem is posed by data quality deficiencies due to imprecise sensors, environmental influences, transfer failures, etc. If not handled carefully, they lead to misguided decisions and inappropriate actions. In this paper, we present the quality-driven optimization of stream processing to improve the resulting quality of data and service. First, we present the optimization objectives and discuss the parameterization of stream processing operators to define the underlying optimization problem. We develop the generic optimization framework and present the quality-driven evolution strategy (QES). Finally, we show that the designed optimization scales very well with regard to processing complexity and reduces numerical errors in the contact lens production monitoring.