Background Neoadjuvant chemotherapy (NAC) followed by radical surgery is a promising strategy to improve survival of patients with stage III gastric cancer, but is associated with the risk of preoperative overdiagnosis by which patients with early disease may receive unnecessary intensive chemotherapy. Methods We assessed the validity of a preoperative diagnostic criterion in a prospective multicenter study. Patients with gastric cancer with a clinical diagnosis of T2/T3/T4, M0, except for diffuse large tumors and extensive bulky nodal disease, were eligible. Prospectively recorded clinical diagnoses (cT category, cN category) were compared with postoperative pathological diagnoses (pT category, pN category, and pathological stage). The primary endpoint was the proportion of pathological stage I tumors among those diagnosed as cT3/T4, which we expected to be 5% or less. Results Data from 1260 patients enrolled from 53 institutions were analyzed. The proportion of pathological stage I tumors in those with a diagnosis of cT3/T4 (primary endpoint) was 12.3%, which was much higher than the prespecified value. The positive predictive value and the sensitivity for pathological stage III tumors were 43.6% and 87.8% respectively. The sensitivity and specificity of contrast-enhanced CT for lymph node metastasis were 62.5% and 65.7% respectively. After exploring several diagnostic criteria, we propose, for future NAC trials in Japan, a diagnosis of “cT3/T4 with cN1/N2/N3,” by which inclusion of pathological stage I tumors was reduced to 6.5%, although its sensitivity for pathological stage III tumors decreased to 64.5%. Conclusion Clinical diagnosis of T3/T4 tumors was not an optimal criterion to select patients for intensive NAC trials because more than 10% of patients with pathological stage I disease were included. We propose the criterion “cT3/T4 and cN1/N2/N3” instead.