A simple model is presented as an aid in understanding, first, the relative noise performance and, second, the noise reduction achievable by referencing, in different experimental approaches to single shot broadband coherent anti-Stokes Raman scattering (CARS). Qualitative agreement is obtained with previous experimental investigations of CARS noise. The broadband dye laser radiation is described as the sum of independent modes with random phases. The dye laser contribution to the CARS noise is then approximately inversely proportional to the square root of the number of dye laser modes generating the detected signal. A fundamental idea is that in Raman resonant spectra only the number of Stokes modes actually participating in driving the Raman resonance should be counted. This means, e.g., that for narrow Raman resonances, as in an atmospheric flame, the noise generated by the dye laser will be higher for a single-mode pump laser than for a multimode pump laser with the experimental CARS configuration normally employed. The implications of the model for the dual broadband type CARS techniques are also discussed.