Thee basically are two different ways of measuring market share that are being thrown around on forums and blogs and those are browser statistics and retail market share. Both of these have their considerable flaws, but still, both are not entirely useless.
Browser statistics, usually those by Net Applications, are flawed because they track only a number of sites. Even though that may encompass 160 million users, it may still be that several target groups do not visit the sites monitored. Still, they do offer an insight into trends. Since the figures by Net Applications are offered monthly, you can see where they are going. Even if they are not good for measuring market share itself, you can still use them to investigate where the market is going: sure, Linux may only have 0.5% according to these figures, but if a year ago this was stuck at 0.25, that still indicates a growing trend.
The other figures usually used in Mac market share are those of retail sales. These are especially problematic, most importantly because they usually always deal with US retail sales alone. No offence to our US buddies, but the US is not the world. Apple may proudly say during a keynote that they have increased sales, but if it's only inside the US, that doesn't tell us much. You can sell a lot more computers outside of the US than inside, and ignoring those figures - as Apple always does - only indicates they would have a very negative impact on the sunny show the company always wants to put up.
Retail figures have other problems, such as the fact that they do not actually measure the amount of computers sold to customers, but only to retailers. Retailers may have a stockpile of unsold computers which still get counted as sales. In addition, retail channels exclude business sales, a segment where the Windows side of things is particularly successful.
What this all means is that you should never trust figures. The funny thing about them is that they are always right when they support your ideas, but always flawed when they do not.