In recent years, oil fire pollution accidents have occurred frequently and caused significant harm. It has become important to study oil fire accidents by extracting fire information through the analysis of spectral characteristics. Many domestic and foreign scholars have established a variety of gas radiation and carbon black radiation models to study fuel combustion; however, few scholars have directly modeled flame spectra to analyze and extract spectral characteristic information on combustion pollution products. In this study, a test platform for the flame spectra of oil was constructed, and the flame spectra of alcohol, 92 gasoline, 95 gasoline, and 0 diesel were measured at a single scale, as was the flame spectra of 0 diesel at multiple scales. The experimental results demonstrate that the flame spectra of the three oils are similar and the radiance increases nonlinearly with an increase in scale. Based on the statistical narrow-band method, a spectral radiation model for the oil flame was established, and a curve fitting degree of 0.895 was obtained based on experimental data. The spectral radiation model can be used to calculate the average radiance and transmittance of oil flames on a large scale, as well as the average transmittance at different flue gas concentrations, which can be helpful for remote fire pollution detection and pollutant concentration inversion.