Both procedure produce descriptive statistics.By proc uni
variate, by default it produce all the statistics(some time
not all required) but in proc means it is possible to
request the statistics that we want.
ex---proc means data=xyz mean max sd;
      run;*it would produce statistics of above which we've
mentioned;

with proc univariate you can genarate histogrmes(graphs) 
for particular variables.you can't with proc means.

PROC UNIVARIATE gives more descriptive statistics such as 
skewness, kurtosis, Q_PLOT and so on.  If you are looking 
for an indepth analysis of the data, like clustering, 
association tree,..., we start with PROC UNIVARIATE.  

If we are looking for some simple results like sum mean SD 
and to find extreme values we use PROC MEANS as it takes 
less of machine time than PROC UNIVARIATE.