Line shifts for some emission lines on the SOHO/CDS NIS detector appear to
be strongly correlated with local intensity gradients along the slit in a w
ay that seems impossible to explain with a physical solar model. Line width
s also show a correlation with local intensity gradients. The most plausibl
e instrumental explanation seems to be an elliptical, tilted point-spread f
unction inducing the line shifts. A toy model demonstrating the essentials
of the observed behaviour is presented. The effective point-spread function
of the instrument appears to modify the line shape into something other th
an a Gaussian, leaving highly structured residuals after line fitting, incl
uding `ghost' images in some pixel planes. The cause of these effects is ye
t unknown, but they should warrant experiments on the engineering model to
reproduce the observed effects, shedding light on the nature of the aberrat
ions.