Determining the best treatment for prostate cancer patients with a newly diagnosed positive biopsy can be challenging. Multivariate prognostic models are often employed to stratify patients into risk populations. Many models leverage quantitative features derived from morphological analysis of the tumor architecture in the biopsy specimen. The vast majority of these features are derived from analyzing standard hematoxylin and eosin (HE) images. Immunofluorescence (IF) image analysis of tissue pathology has also recently been proven to be robust. In this work, we constructed multivariate models for prostate cancer prognosis comparing the usage of previously published IF vs HE features. In images from 304 patients, the IF features prognostically outperform the HE features. The IF feature model also exhibits consistent training vs validation performance, an important consideration when developing models subject to regulatory oversight. This paper presents the first evaluation of comparing previously published HE and IF morphological features head-to-head in prognostic models from prostate biopsies.