According to Rundle's curve, Graves' ophthalmopathy (GO) worsens during an initial phase up to a peak of maximum severity, then improves and reaches a static plateau, with the activity curve preceding the severity curve by a few months. To our knowledge, no studies have tried to replicate Rundle's curve, and very few have investigated the natural history of GO. Here, we studied GO natural history retrospectively and tried to identify factors that may affect it. A total of 65 patients with untreated GO underwent an eye assessment after a median of seven months after the appearance of GO and then after a median of 40 months. The primary endpoints were the variation of the single GO features and of the NOSPECS score, as well as the overall outcome of GO. The secondary endpoint was the influence of several variables (age, sex, smoking, GO and thyroid disease duration, thyroid treatment, thyroid status, thyroid volume, anti-TSH receptor autoantibodies) on the outcome of GO. The majority of patients had mild, minimally active GO, and only five had a Clinical Activity Score (CAS) >3. There was a significant reduction of CAS (p<0.0001) and NOSPECS (p=0.01) between the first and last observation, with a timing pattern resembling Rundle's curve. This difference was confirmed even when patients with a CAS >3 at first observation were excluded. At the last observation, 50.8% of patients had improved, 33.8% had remained stable, and 15.4% had worsened moderately or substantially. The overall outcome of GO was not affected by any of the variables under examination. In confirmation of Rundle's observations, untreated GO improves spontaneously with time in the majority of patients, with an activity peak between 13 and 24 months, which may have implications in determining the proper timing of GO treatments.