Something I wish we had done better is base lining story points.
Our team, and the organization we work for had always had lot's of trouble estimating. Planning poker is the tool that has at least partly increased the teams ability to estimate the work. Unfortunately the estimates still give me plenty of grief. That is the estimates are regularly but inconsistently out by significant amounts.
There are many issues at play here; and many of them are typical things reported in the scrum and agile communities. Main problems have included seeking too much information and seeking too much precision. A third problem is a drift in the baselines.
What we did right:
We took the first couple of dozen stories and estimated them by the Mike Cohn book. Out of that session we took a handful of baseline stories as put them on the wall as reference markers for future estimating.
What we did wrong:
Eventually we took down the baseline stories so we could use the white board for other things. The team lost its reference points.
(Another important problem was that we (not everyone, but enough) forgot the purpose of estimates is to estimate, not to solve the feature design, but that's another story and one that has been well covered elsewhere.)
Put the baseline stories on the wall and leave them there. Always have them in line of sight from where you do your estimates.