I want to develop an inequality analisys and evaluate Gini index at county level on personal data. I would expect to see differences among different counties. Should I use sample weights or are they evaluated with reference to the national level?
The ACS samples (available since 2000) include weights that are adjusted for geographic sampling rates, nonresponse, and individual sampling probabilities. They are intended to be representative of the geographic area of interest, so you should use the same weights for the county level as you would for the national level. You can find further information on the ACS sample design and weighting here, as well as documentation of non-ACS samples.
Additionally, you can find more information on the construction of weights at the IPUMS FAQ, the Accuracy of PUMS documentation, and the PERWT variable description (includes additional links). These should help guide you in your use of weights at subnational levels, such as county.
One other potentially helpful resource is American FactFinder. You can search for official Census summary tables of subnational regions (e.g. county, MSA, etc.) and compare these numbers against your own subnational estimates using IPUMS data. Since PUMS data is a subset of the data used in the Census summary tables, your estimates are unlikely to match exactly; however, you should still be able to produce similar results using the PUMS sample weights.
I hope this helps.