A study is made of the intensity of the Carrington magnetic storm of September 1859 as inferred from visual measurements of horizontal-component geomagnetic disturbance made at the Colaba observatory in India. Using data from modern observatories, a lognormal statistical model of storm intensity is developed, to characterize the maximum-negative value of the storm-time disturbance index (maximum –Dst) versus geomagnetic disturbance recorded at low-latitude observatories during magnetic storms. With this model and a recently published presentation of the Colaba data, the most likely maximum –Dst of the Carrington storm and its credibility interval are estimated. A related model is used to examine individual Colaba disturbance values reported for the Carrington storm. Results indicate that only about one in a million storms with maximum –Dst like the Carrington storm would result in local disturbance greater than that reported from Colaba. This indicates that either the Colaba data were affected by magnetospheric-ionospheric current systems in addition to the ring current, or there might be something wrong with the Colaba data. If the most extreme Colaba disturbance value is included in the analysis, then, of all hypothetical storms generating the hourly average disturbance recorded at Colaba during the Carrington storm, the median maximum –Dst = 964 nT, with a 68% credibility interval of [855,1087] nT. If the most extreme Colaba disturbance value is excluded from the analysis, then the median maximum –Dst = 866 nT, with a 68% credibility interval of [768,977] nT. The widths of these intervals indicate that estimates of the occurrence frequency of Carrington-class storms are very uncertain, as are related estimates of risk for modern technological systems.