I've working on cholesterol quantitation in food samples, specially seafood. I've noticed wide differences among injections of a same extract or between replicates of a same sample. This behavior is not constant, given that sometimes a good repeatability is shown. After cleaning the detector (FID), injector and septum replace, this appears to get better but not for much time. In a work day, it becomes worse and worse. The day after I can have a good response at the begining, and the cycle goes on.
Some people advise me to take care with split/splitless injector, since this causes great variability, in particular if manual injection is used. What can I do for repeatability performance?

We assayed a lot of TMS derivatives of various sterols on a HP5890 system and never had a problem. The matrix was sediment samples, so it was not exactly clean. We used to replace the inserts (pre-silylated, Supelco sell a nice premixed soln.) regularly and nip a bit of the front of the column. I cannot ever remember cleaning the FID more regularly than every month.

I do cholesterol analysis, of oils, normally I have no problem with replication between injections. I use an on-column injection, though. Do you do any clean up stage?

I don't know if this was the cause but, since the flow valve was replaced, I don't have problems with repeatability (until now).
About clean up, I filter the samples through glass wood. The problem I think is that some of the samples are freeze dried, then the components are very concentrated. When I store the extracts of that kind of samples into the freezer, a sediment develops so I have to centrifuge them and inject the supernatant.