Guided wave (GW) has been widely used in engineering structures for structural health monitoring (SHM) because of its capability of long propagation and high sensitivity to local defects. Bayesian methods have been commonly employed to enhance the reliability of GW-based damage identification. In addition to more accurate detection, the corresponding uncertainties within unknown parameters can be quantified by assessing the posterior distributions. However, applications of most existing Bayesian frameworks require a proper likelihood assumption, which tends to restrict the applicability when dealing with a complex or high-dimensional problem. Conversely, approximate Bayesian computation (ABC), a likelihood-free method while still obeying the Bayes theorem, provides a more straightforward alternative to solving the Bayesian inference problem. The posterior distributions of unknowns are estimated using a distance function to assess the similarity between simulation and measurement. Summary statistics, such as signal processing techniques in GW, can also be used to pre-process datasets to enhance comparison sensitivity. In these cases, no explicit likelihood function is needed for posterior estimation. In this paper, A GW-based damage identification framework is proposed using ABC. An experiment of multiple cracks in the isotropic metal rod is carried out. The measurement is used in the proposed ABC framework to detect the corresponding damage locations and sizes. Extensive studies are conducted to evaluate the ABC performance. Recommendations are provided based on identification results. Additionally, this research provides a further understanding of ABC framework development for GW-based SHM.