Ok, link to what other people have written first - Good find Slevin btw.

I did a search on the forum using the terms:
confidence AND deviation AND 95%

This brought up 16 links which are probably all worth reading for deep insight into this issue. I particularly like this one:
http://www.flopturnriver.com/phpBB2/...n%20confidence

Now, to blurb a few thoughts out.

When talking about win rates it's important to realise that everything has a foundation - and the foundation for people speculating in and evaluating their win rate is historical. Specifically the calculations are ones used by poker players who made their living for many years prior to the poker boom by playing limit holdem. In limit holdem they have a small blind, a big blind, a small bet (equal to the big blind) and a big bet (equal to double the big blind). Pre-flop and flop you make small bets (fixed size), and on turn and river you make big bets (fixed size).

People decided then (for whatever reason) that they wanted to calculate their win rates in a number relative to the standard blinds or bets - and they decided to use the big bet. If you play limit I'm sure this makes perfect sense, and if you play no limit it can be more than a little confusing if you don't know why this is the case.

Particularly - PokerTracker records win rate in bb/100 - which means big bets per 100 hands. In NL there are no big bets as a predefined fixed size - what it means is the amount that a big bet would be, if it was a limit game with these blind sizes. It's a convention and one we really just get used to.

To make it absolutely clear that we're not talking big blinds when we say bb/100 it is customary (or at least common) to say ptbb/100 - this abbreviation means pokertracker big bets and is used mainly to get people asking if you are talking about big blinds or something else.

Now, I play 10k hands, I win 1000 big blinds (10 buyins) - what's my win rate? Well, 10k hands is 100 times 100 hands - so we need to divide by 100 to get a win rate per 100. That's 10 big blinds per 100. How many ptbb is this? That's 5 ptbb/100. Generally if people say 5 ptbb/100 or 5bb/100 they tend to mean the same thing (but it's ok to reality check their math or just plain ask if in doubt).

In a sense this is only the mere beginning of establishing your win rate. I may win some today and lose some tomorrow and my overall win rate for the two days is.. something. I may win or lose small or big amounts in any given session. My past performance gives an indication what my win rate is, but what if I've just been lucky? Or unlucky?

There are two main tools we begin to apply when determining a more true and better picture of a win rate.

1) Determine your luck. There are tools (graphing tools and others) that come out with numbers like Sklansky bucks and set-o-meter stats which provide a kind of rough numbers to indicate how lucky you've been with your cards (and may suggest what your win rate would have been if you had completely neutral luck with cards).
2) Determine with a degree of mathematical certainty what your win rate is. Particularly the thread I listed above goes (in some posts) into discussing variance. Every session you play may be winning or losing, and what's important is not just your average win over time, but also how big the variance can be in a given session. There's a family of math techniques that can take my above 5 ptbb/100 example apart, judge all the variance involved and state that since my sample size is small (10k hands is small) and my variance per session / hour / 100 hands is big it is mathematically 95% likely that my true win rate is between 1 and 9 ptbb/100.

This is called the 95% confidence interval (and informed my search criteria above).

What I've done here is not even a basic introduction to how to do it, but just an overview type description to try to give a basic understanding what the techniques can and cannot do.