The purpose of this study is to quantify the discrepancy between optical guidance platform (OGP) frameless localization system (Varian) and Trilogy on-board imaging (OBI) system (Varian) for setting up phantom and stereotactic radiosurgery (SRS) patient; and to determine whether cone-beam CT (CBCT) is necessary for OGP patient setup, and compare CBCT and orthogonal kV-kV in term of their verification capability. Three different phantoms were used in the study: a custom-made phantom, a Penta-Guide phantom, and a RANDO phantom. Five patients using both OGP and CBCT setup and 14 patients using CBCT setup alone were analyzed. One patient who had big couch shifts discrepancy between OGP and CBCT was selected for further investigation. Same patient's CBCT and planning CT were fused. A RANDO phantom simulation experiment was performed using OGP setup with both CBCT and orthogonal kV-kV verification. For all of three phantom experiments, the shifts performed by CBCT beam and orthogonal kV-kV were all within 1 mm. Among five SRS patients using OGP setup, three had 3D couch corrections more than 3 mm. The image fusion of CBCT and planning CT clearly illustrated a tilt of bite-block in a patient's mouth. For 14 SRS patients using CBCT-guided setup, overall 3D correction was 3.3 ± 1.5 mm. RANDO phantom experiment demonstrated how a tilted bite-block caused isocenter shift. CBCT-calculated shifts are the same as expected, but kV-kV results differed by 1-2 mm if the initial head position is tilted. The bite-block tilting in patient's mouth is a major reason for the cause of positioning error for OGP frameless SRS setup. CBCT verification is necessary. CBCT provides more accurate couch corrections than orthogonal kV-kV when head was tilted. OGP is useful for detecting patient movement, but it does not necessarily imply that the isocenter has moved.