Both programs use the same correlation matrix? Lisrell use polichorich if items are ordinal, it could be the cause?

I think that I found a possible explanation:

"The user should use some caution, because programs do not always recalculate incremental fit indices such as the CFI, TLI, or the IFI using the scaled chi-square for the tested model or the null model (I know that Mplus and EQS do use the scaled chi-squares in their calculation).  Relative fit indices will likely be problematic when scaling corrections to the null model are not used (Hu & Bentler, 1999)."

(Newsom, J. (2005). Practical approaches to dealing with nonnormal and categorical variables. Retrieved October, 9, 2008.)

You might consider posting this question to the discussion board on the Mplus website (http://www.statmodel.com). The program's developers (Bengt and Linda Muthén) check the discussion boards quite frequently.

Noelia, the reason for your problem may be due to different formulas used for the correction, LISREL uses the Satorra-Bentler correction for MLR:

www.ssicentral.com/lisrel/techdocs/ftb.pdf

However, the Satorra-Bentler correction used by LISREL is different to the correction used by Mplus, this is the Yuan-Bentler correction (see the Mplus manual). The differences are often quite small, but sometimes they are large. This is explained in Yves Rosseel's presention on the comparison between Mplus and EQS:

http://users.ugent.be/~yrosseel/lavaan/utrecht2010.pdf

Hope this helps!

Regards, Karin

Is it possible that differences also exist for RMSEA reported by LISREL and MPlus?