I believe you just look at the absolute value of the t statistic. The sign will change depending on what group you call group 1 or group 2. So there's no contradiction; you should reject the null. Just out of curiosity, though, note that although you assume equal variance, the variances in the two groups are not equal. I doubt that will cause a problem big enough to make you unable to reject the null, but you can do a t-test that accounts for unequal variances - I think it is an option in excel.

I did a fischer test on the variance, and there isn't a significant difference between them. The absolute value eh.. interesting!!! excel lies!!!

I'm not sure Excel exactly lies, but it's not a great program for statistics. I'm a retired infectious disease epidemiologist; I used to use the program Stata for statistics. It's pretty easy to use (menu-driven) and it covers almost everything you're likely to need unless you are doing very complex analyses on huge databases. I've heard SPSS is good for social science statistics, but it's not quite as good for epidemiology as Stata. The monster was always SAS - that was driven by commands without a menu and folks who learned it were very attached to it, but unless you needed to do something very tricky, I don't think it was any better than SPSS or Stata.