Iterative CT reconstruction algorithms are gaining popularity as GPU-based computation becomes more accessible. These algorithms are desirable in x-ray CT for their ability to achieve similar image quality at a fraction of the dose required for standard filtered backprojection reconstructions. In optical CT dosimetry, the noise reduction capability of such algorithms is similarly desirable because noise has a detrimental effect on the precision of dosimetric analysis, and can create misleading test results. In this study, we evaluate an iterative CT reconstruction algorithm for gel dosimetry, with special attention to the challenging dosimetry of small fields. An existing ordered subsets convex algorithm using total variation minimization regularization (OSC-TV) was implemented. Three datasets, which represent the extreme cases of gel dosimetry, were examined: a large, 15cm diameter uniform phantom, a 1.35cm diameter finger phantom, and a 15cm gel dosimeter irradiated with 3×3, 2×2, 1×1, and 0.6×0.6cm fields. These were scanned on an in-house scanning laser system, and reconstructed with both filtered backprojection and OSC-TV with a range of regularization constants. The contrast to artifact+noise ratio (CANR) and penumbra width measurements (80% to 20% and 95% to 5% distances) were used to compare reconstructions. Our results showed that OSC-TV can achieve 3-5× improvement in contrast to artifact+noise ratio compared to filtered backprojection, while preserving the shape of steep dose gradients. For very small objects (≤0.6×0.6cm fields in a 16×16cm field of view), the mean value in the center of the object can be suppressed if the regularization constant is improperly set, which must be avoided. Overall, the results indicate that OSC-TV is a suitable reconstruction algorithm for gel dosimetry, provided care is taken in setting the regularization parameter when reconstructing objects that are small compared to the scanner field of view.