CFI is often much friendlier than Chi-square and RMSEA. I would consider the sample size (which you don't give). If it is very large (several thousand cases) I can imagine going with CFI. At least if the correlations are not very low in which case CFI can mistakenly point to good fit.
I would also say that the chi-square and RMSEA results indicate that it is possible to modify your model to get better fit. You can check this out using Modindices.
Konstantin posted on Thursday, October 16, 2014 - 2:24 am
yes i have big data set, more than 1000 observations here its https://www.sendspace.com/file/z4ayk4 i have uploaded on web share, cause i can't see how attache file here. programm could work with variables X4-X6 so they were excluded