Pipelines constitute crucial infrastructure in the oil, gas, chemical, and water transport industries. In-service pipelines are prone to a variety of defects that stem from the deleterious effects of fatigue, aging, external impacts, and corrosion from hazardous operating environments. There is thus an urgent need for inspection techniques that can detect pipeline defects and characterize those defects to allow maintenance and replacement operations to be carried out accurately and efficiently. The ultrasonic guided-wave technique is an active inspection approach that has undergone recent evolution and demonstrated great potential for the nondestructive testing of materials and structures in a variety of fields. However, defect characterization in guided wave-based pipeline inspection remains extremely challenging because of the complexities involved in the interaction between guided waves and pipeline defects, and the resulting reflection signal that conveys multidimensional geometrical information in those defects. In this work, the reflection-related problems of the guided waves from pipeline defects are first investigated through extensive laboratory experiments and numerical simulations in parallel with specific concerns about the reflections at the defect edges. The results of this investigation show the reflection from a defect to be the joint result of the interference between the reflections at its front and back edges. The edge reflection components embedded in the defect reflection signal exhibit different features, thereby further increasing the complexity of the total defect reflection signal. The relationships between each edge reflection and the three-dimensional geometrical parameters of the defect, including its axial length and circumferential extent and radial depth, are identified. The results indicate that the pattern of the reflection waveform from a pipeline defect is affected primarily by the defect's axial length, whereas the reflection amplitude is strongly determined by the geometrical parameters of its edges. The findings presented herein provide new perspectives and offer useful guidance on the interpretation of the reflection signals in guided wave based pipeline defect inspection. Based on this foundation, a new strategy for characterizing pipeline defects is formulated. The framework of this strategy requires that the edge reflection components first be decomposed from the defect reflection signal and then analyzed to enable the quantitative characterization of defect. The applicability of the new characterization strategy to arbitrary pipeline defects is explored semi-analytically using transmission-line model technique. As the primary components of defect reflection signal, the edge reflection signals are generated from partial reflection defect edge with respect to the entire edge, which is termed as effective edge and closely related to the critical geometrical features of the defect. The characteristic dimension of pipeline defect is further employed to describe the results obtained under the framework of the new strategy. This strategy reduces the complexity of analyzing the defect reflection problem because it removes the destructive effects of the interference between the reflection components. Moreover, it provides additional defect geometry-related information sources, thereby enhancing the reliability and accuracy of quantitative defect characterization. In implementing edge reflection-based characterization strategy, the problems of signal decomposition and analysis remain. Two methods of decomposing the defect reflection signal into edge reflection components are developed in this thesis. These methods permit a determination of the characteristic axial length of the defect, which corresponds to the relative distance between two decomposed edge reflections. The guided wave reflection at the edge of defect is also investigated under different defect cases, with the results showing that the extent of the F(1,3) mode generated during the edge reflection due to mode conversion is comparable to that of the incident L(0,2) mode, which can provide circumferential information of the defect. The relationship between edge reflection and the geometrical parameters of defect edge is further constructed using the least squares support vector machine method, the results of which can be used in conjunction with the circumference extent to determine the radial depth. Various types of defects, including simulated notches, artificial notches, and real pipe corrosion, are considered to demonstrate the capabilities of the methods developed herein.