This work presents a methodology for inferring a fire's radiative fraction from an ensemble of heat flux measurements. For each heat flux measurement, the simple point source model is used to infer the radiative heat release rate. A set of calibration experiments allows for the characterization of the systematic (bias) and random errors associated with the approach. An exploration of the systematic errors shows that flame leaning likely due to ventilation effects leads to overestimating the radiative heat release rate for sensors in the direction of the flame leaning and underestimating the radiative heat release rate for sensors that are opposite the direction of the flame leaning. After correcting the systematic errors in the point source model's predictions, the random errors are also characterized using the calibration experiments. Rather than assuming that the errors are independent, a covariance matrix is computed, which accounts for the statistical dependence between point source predictions. Because the random prediction errors are not statistically independent across different sensors, there is diminishing utility in adding additional sensors. The analysis shows that using four heat flux sensors produces nearly the same uncertainty as using eight. Finally, the statistical characterization of the radiative heat release rate provides a framework for the inference of the radiative fraction when combined with measurements that allow for inference of the total heat release rate.