This thesis addresses the problem of source quality changes in Data Integration Systems. Its main goal is to propose techniques for maintaining as much as possible the satisfaction of users' quality requirements. It proposes an approach that bases on a proactive strategy, where probabilistic techniques are applied. These techniques allow to model source quality behavior, and to calculate quality reliability of the system. On the other hand, it proposes a reactive strategy that must be applied for compensating source quality changes. Its detailed solutions focus on freshness and accuracy quality properties.