Background: The diagnosis of COVID-19 based on clinical evaluation is difficult because symptoms often overlap with other respiratory diseases. A clinical score predictive of COVID-19 based on readily assessed variables may be useful in settings with restricted or no access to molecular diagnostic tests. Methods: A score based on demographics and symptoms was developed in a cross-sectional study including patients attended in a dedicated COVID-19 screening unit. A backward stepwise logistic regression model was constructed and values for each variable were assigned according to their β coefficient values in the final model. Receiver operating characteristic (ROC) curve was constructed and its area under the curve (AUC) was calculated. Results: A total of 464 patients were included: 98 (21.1%) COVID-19 and 366 (78.9%) non-COVID-19 patients. The score included variables independently associated with COVID-19 in the final model: age equal or above 60 years (2 points), fever (2), dyspnea (1), fatigue (1 point) and coryza (-1). Score values were significantly higher in COVID-19 than non-COVID-19 patients: median (Interquartile Range), 3 (2-4), and 1 (0-2), respectively; P<0.001. The score had an AUC of 0.80 (95% Confidence Interval [CI], 0.76-0.86). The specificity of scores equal or greater than 4 and 5 points were 90.4 (95%CI, 87.0-93.3) and 96.2 (95%CI, 93.7-97.9), respectively. Conclusions: This preliminary score based on patients symptoms is a feasible tool that may be useful in setting with restricted or no access to molecular tests in a pandemic period, owing to the high specificity. Further studies are required to validate the score in other populations.