kreno,
it would take about two weeks to become familiar enough to begin to be able to play w/ datasets. that assumes that you have a Linux machine already setup w/ AFNI and some scans to play w/. prolly a month after that to produce something of interest (assuming that you have subjects and data collected).
if you are starting from scratch or you are at a center w/ advanced analysis in fMRI data going on, i would suggest learning freesurfer (an MGH/MIT overlay on AFNI that should allow much more precise data analysis) along w/ AFNI basics.
if you are new to the principles of functional imaging, i would suggest reading some papers or chapters on the fundamentals of functional brain imaging, and the principles and drawbacks of BOLD and other fMRI techniques and block vs event-related design, etc. better yet, if you can get funded, i would suggest taking an fMR workshop class by the likes of MCW, MGH, UCLA or the NIH (also Wash U, Stanford, U Pitt, U Penn, U Chicago (to some extent), among other leaders in the field)
also check out the AFNI documentation pages and the AFNI message board. Dr. Robert Cox, the "godfather" of AFNI is great at recognizing the advantages and disadvantages of his techniques.
http://afni.nimh.nih.gov/afni/
above all, as you master the basic of computer analysis, understand that the most important first step is to design and execute experiments that actually address the question being asked. there's not much more frustrating than chasing an epiphenomenon or ghost of poor initial experimental design.