Confirmatory Factor analysis

#1
Perhaops this question is a little too advanced for this forum, but I will give it a try.

I am doing CFA in AMOS and I am getting rather bad fit indices. It look like the "problem" is that two of the items are highly correlated. (I have only one factor btw.)

Now if I put a double-headed arrow between the two items then the fit-indices get better. In other words the model better fits the data. However, I know that the items are correlated. I would be astounded if they were NOT! So it looks like I have somehow assumed (in the set up of the CFA) that the items are uncorrelated - or is that a standard assumption? Because conceptually I definitely assume they ARE correlated.

Can I fit a CFA model that automatically assumes some level of correlation between the items?

___

Any other comments?