Purpose: This study was done to examine predictive validity of Critical Care Non-verbal Pain Scale (CNPS) and develop criteria for pain assessment using CNPS with critically ill patients who have communication problems. Methods: Data were collected from intensive care units at three major general hospitals in Seoul and Kyunggi province. During each observation, a nurse assessed pain severity using CNPS ratings (range 0-9) at four treatment stages: at rest, during central catheter dressing change (nonpainful procedure), position change and suctioning (routine painful procedures). Patients also assessed their pain using a self-report 4-point VRS-4. Results: There were significant differences between the four treatment stages except between "at rest" and "nonpainful procedure". Strong correlations were found between CNPS and VRS-4 for "at rest" (r=.552, p<.001), central catheter dressing change (r=.505, p<.001), position change (r=.709, p<.001), and suctioning (r=.662, p<.001). ROC curve analysis of CNPS based on 3 point on VRS-4 showed the cutoff point was 3 for CNPS, the starting point for pain management with 73% sensitivity, 92.2% specificity, 73% positive predictive value, and 92.8% negative predictive value. Conclusion: Results indicate that CNPS is a valid tool for measuring pain in critically ill patients with communication problems and 3 point should be the standardized pain treatment point.