Narcissistic ghost lines in a Rowland-grating spectrograph have been identified as arising from the reimaging of bright spectral features, which are incident on the face of a detector photocathode [Cesium iodide (CsI) on a microchannel plate], back onto the detector by the grating in zero order. The mean of the wavelength of the diffracted light and the apparent wavelength of the ghost allows the angle of the grating normal with respect to the input beam (alpha) to be determined. Measurements of ghost intensity as a function of wavelength are presented and are found to range between 7 x 10(-4) and 7 x 10(-3) of the parent line. We find that the sum of the CsI photocathode reflectivity and quantum efficiency <1/2, showing the bulk of the light incident upon the detector, is neither reflected nor detected. We caution that any Rowland circle spectrograph with a detector normal nearly aligned with the grating normal and with a sufficiently reflective detector face (or surrounding mounting structure) will suffer from these narcissistic ghosts.