We have measured the manganese abundance in 24 HgMn stars using a
curve-of-growth technique, using 6 well characterised Mn I and Mn II
lines. A more detailed spectrum synthesis analysis for a selection of
stars confirms the accuracy of this simpler method when a few good
unblended lines of moderate strength are used. The results agree well
with the UV abundances of SD93 for all the isolated stars and some of the
binaries. The remaining binary stars show a higher abundance in the
visible region than the UV, which is to be expected, since SD93 did not
account for binarity. The remarkably small scatter amongst the abundances
derived for individual lines, and the excellent agreement with the UV
results, show that careful selection of a few well characterised lines is
a powerful tool for a quick and accurate abundance analysis. The
similarity between the abundances derived for Mn I and Mn II may imply
that we see very little in the way of non-LTE effects. The results support
the comparison by SD93 of the predictions of diffusion theory by Alecian
& Michaud [1981] with abundance trends for the upper envelope
of abundances as a function of 
. This is another detailed prediction
of the theory which has been confirmed observationally.
We have presented strong evidence that hyperfine structure is the cause of
the widened line profiles and is mainly responsible for the line-strength
anomalies seen in Mn II 
4326
and 4206, and we have shown
that hfs can be a major source of systematic error if not taken into
account in abundance analyses. The other Mn lines studied must have very
much narrower hfs structures, to give such consistent results. We were
able to derive crude ad hoc structures, but a laboratory study of
hfs in Mn II is urgently needed if more precise results are to be
obtained.