Purpose: The pelvic phantom was fabricated in the following purposes: (1) Dose verification of IMRT plan using Eclipse planning computer, (2) to study the interface effect at the interface between rectal wall and air. The TLD can be inserted in the pelvic phantom to confirm the dose distribution as well as uncertainty at the interface. Materials and Methods: A pelvic phantom with the dimension of 30 cm diameter, 20 cm height and 20 cm thickness was fabricated to investigate the dose at the rectal wall. The phantom was filled with water and has many features like bladder, rectum, and prostate and seminal vesicle (SV). The rectum is made of 3 cm-dimater plastic pipe, and it cab be blocked by using a plug, and film can be inserted around the rectal wall. The phantom was scanned with Philips Brillance scanner and various organs such as prostate, SV, and rectal wall, and bladder wall were delineated. The treatment parameters used in this study are the same as those used in the protocols in the SNUH. TLD chips are inserted to the phantom to evaluate the dose distribution to the rectal wall (to simulate high dose gradient region), bladder wall and SV (to simulate the high dose region) and 2 spots in anterior surface (to simulate the low dose region). The TLD readings are compared with those of the planning computer (ECLIPSE, Varian, USA). Results: The target TLD doses represented as the prostate and SV show excellent agreements with the doses from the RTP within +/-3%. The rectal wall doses measured at the rectal wall are different from the those of the RTP by -11%. This is in literatures called as an interface effect. The underdosages at the rectal wall is independent of 3 heterogeneity correction algorithm in the Eclipse RTP. Also the low dose regions s represented as surface in this study were within +/-1%. Conclusion: The RTP estimate the dosage very accurately withihn +/-3% in the high dose (SV, or prostate) and low dose region (surface). However, the dosage at the rectal wall differed by as much as 11% (In literatures, the underdosage of 9$sim$15% were reported). This range of errors occurs at the interface, for example, at the interface between lung and chest wall, or vocal cord. This interface effect is very important in clinical situations, for example, to estimate the NTCP (normal tissue complication probability) and to estimate the limitations of the current RTP system. Monte-carlo-based RTP will handle this issue correctly.