Auroral roar is an electromagnetic emission near two and three times the electron gyrofrequency. The generation mechanism is believed to involve mode conversion of upper hybrid waves generated where the upper hybrid frequency matches a cyclotron harmonic and trapped in density enhancements. Theory predicts that if the density enhancement has dimensions comparable to the wavelength, the waves will be discretized, and indeed discrete features have been observed with the electric field detector on the HIBAR rocket experiment and with ground-level instruments. We have generated a database of eigenfrequencies predicted by theory as a function of the dimension and amplitude of the trapping density enhancement, assuming a Lorentzian-shaped cyclindrical trapping density enhancement embedded in the auroral F-region ionosphere. A fitting method is developed to test these predictions against the rocket and ground-based data. The fitting shows a significant discrepancy between theoretical predictions of density profiles and HIBAR observations. The predicted density enhancement scale size is of order ∼10 m, whereas rocket observations suggest scale size of order ∼100 m. A correction factor of 30 is used to reconcile the difference between the theoretical predictions and observation. A Versatile Electromagnetic Waveform (VIEW) receiver, operated at South Pole station during 2003, recorded many examples of auroral roar fine structures. Twenty-one relatively stationary examples of fine structures were selected, and the fitting procedure was applied to them, resulting in statistics of the trapping density enhancements associated with them, assuming that the eigenmode theory applies. In this case, the eigenmode theory indicates that irregularities in the range 7–40 m, or 200–1200 m if the correction factor is applied, would be responsible for generating observed auroral roar fine structures.