What's the source of the data? Is it integer, or might you actually have some small fractional parts that are unequal, but are close enough together that the values display as equal? (And is there anything odd going on in the formats for these variables?) That might explain the multiple "equal" rows from PROC FREQ, and perhaps even why FREQ and MEANS give different results -- though not the dramatic differences in values that you show!
I like the earlier suggestion of working with a small subset to figure out what's going on; I'd add a suggestion that you standardize all formats to something with enough precision to tell in detail what you've got, and perhaps even explicitly test whether you have small hidden fractions by testing whether x=floor(x).
>Now, the problem is that the data being counted doesn't merely appear
>the way PROC FREQ counts it. In fact, when looking at the data (having
>saved it to a permanent record a couple times), I would see several
>identical rows together, each with a COUNT of 1; these are rows that
>should have been combined.
Director of Biostatistics
Spectrum Pharmaceutical Research Corp.
San Antonio, TX
SAlbert at SpectrumCRO dot com