One of the objectives of a cancer registry is to provide survival information on subsections of the population that have unfavorable outcomes. A cancer control strategy can be planned on the basis of such information. In the present study, the data of the Songkhla Cancer Registry were analyzed to determine if social and geographic parameters can be used to predict cancer survival. A total of 3423 cases identified in the population-based cancer registry of Songkhla Province registered during 1990-1994, were the subjects of this study. The rest were excluded because of unknown primary cancer sites. Eight leading primary cancer sites were focused on: oral cavity, pharynx, esophagus, colorectum, lung, liver, breast, and cervix uteri. Predictors of survival were derived from items recorded in the registry. Age, gender, extent of disease, cultural belief, life-style, and access to medical care were the predictors of interest. Religion, urban environment, and distance to tertiary care centers were proxies for the last three parameters. Kaplan-Meier plots, Cox regression, and log-rank tests were used for analysis of the hazard ratios. The results revealed a significance of disease extent for survival from oral, colorectal, breast, and cervical cancers. Muslim people had poorer survival rates than those of Buddhists for oral, breast, and cervical cancers. Women with breast cancer living in distant from tertiary medical care centers had a poorer prognosis. For the non-aggressive cancers, early detection and pretreatment counseling for the prevention of unnecessary incomplete treatment is recommended for prolonged survival. With aggressive cancers such as lung, liver, and pharyngeal cancers, there were no significant differences with these parameters investigated. Good quality of life provided by palliative care, not prolongation of survival, is the ultimate goal of medical care services to such patients.