Measuring irradiance just beneath the air-water interface Ed(0-), is challenging because of environmental variability of the incident radiation field, such as effects of waves, perturbation by the instrument platform and instrument limitations. Accurate measurements of subsurface irradiance and radiance, however, are critical in the estimation of remote-sensing reflectance values and the development of ocean color algorithms. Subsurface irradiance is typically estimated by extrapolating measured near- surface underwater spectra back to just beneath the surface. Such an approach, assumes that the water's optical properties are consistent within the extrapolation interval. However, the diffuse attenuation coefficients can vary widely in the surface layer due to selective absorption of the short and long wavelengths, pigment concentrations, and ship shadow effects and are strongly dependent on the sampling depth used in the calculation. Another independent estimate of Ed(0-) is derived by propagating irradiance measured above the sea surface to just beneath the air-water interface. Here, we compare these two independent estimates of Ed(0-) to examine the accuracy of our methods and instrumentation. We use measurements of downwelling spectral irradiance collected over two seasons at Palmer Station, Antarctica using a Profiling Reflectance Radiometer deployed in freefall mode from a small zodiac, so as to minimize ship shadow effects. While estimates of Ed(0-) made from above and below the sea surface data wee highly correlated for overcast days, clear days showed much more scatter between the two estimates. This was attributed to wave effects and the lack of completely clear skies without haze or high clouds. Comparison of above and below water observations with theoretical computations suggest systematic error in immersion coefficients used to calibrate the instrument. Further, very shallow density structure introduces layers of water with differing optical properties and causes error in the estimation of Ed(0-).