Calibration refers to the relationship between predicted and actual results over the range of possible probability categorizations.  It measures the "goodness of fit" or the trustworthiness of the data.
 

Calibration can be assessed in several ways.  Often a table with predicted versus actual results may be provided for inspection.  For example, the following table is calibration data for the pneumonia severity index.

Or, it may be presented graphically

One more example

Finally, calibration information may be expressed statistically, usually with the Hosmer-Lemeshow goodness-of-fit statistic.