This thesis is about early detection of requirements defects. Software-centred systems’ defects can cause loss of life, loss of property, loss of data and economic losses. Requirements defects are a major source of system defects. The early detection of requirements defects prevents software-centred systems’ defects, and thus reduces the various types of losses. In the past thirty years, many methods have been developed to detect requirements defects. The most prominent methods include inspections, automated static analysis, simulation, formal specifications and more recently model-checking. Each method has different strengths and weaknesses. The lack of integration of the different detection techniques produces a knowledge gap that causes problems with repeatability, scalability, effectiveness, and efficiency of the detection process. This knowledge gap is enlarged by the lack of a well-specified defect classification scheme that specifies quality rules, collects defects, specifies defect patterns, and classifies the patterns. This thesis proposes a framework for early defect detection based on Behavior trees, a representation which makes it practical to integrate the various detection techniques. Individual requirements are translated one at a time into Requirements Behavior Trees. These Requirements Behavior Trees are then integrated into an Integrated Behavior Tree that can be inspected, statically analysed, model checked and simulated. The framework is based on the hypothesis that if a well-specified defect classification scheme is developed and different types of detectors are integrated to detect patterns that suit their capabilities and if processes are developed to cover the complete requirements lifecycle, then the framework’s detection results will be more effective and more efficient, and the results will be more repeatable and scalable than existing methods. The framework includes a Behavior Trees defect classification scheme. The scheme defines defect patterns for requirements written in English and requirements specified by Behavior Trees. The scheme has a variety of defect patterns. Each defect pattern contains the characteristics of a type of defect. Defect patterns are grouped together based on the quality rules that they violate. This framework and the hypothesis have been tested using four case studies. The results of the case studies found that compared to the Perspective-based Reading method and three conventional requirements analysis methods the framework proposed was more effective and able to detect a broader range of defect types. However, because of the lack of the tool support, the efficiency of the method is still questionable. It should however improve with better tool support.