It all depends on the number of messages you get.
Most people are reporting a very good accuracy ( > 97% ) after some 1000 messages. The global statistics (see link below), tell us that after 500 messages the average accuracy is at 98.29% and that over 92% of users are getting an accuracy over 95% at that point (as of 28 September 2009; check the link below for the latest data).
This does not mean that you will have to reclassify a thousand times before you reach that point. You will see that the majority of those 1000 messages will be put into the correct bucket. But you should be prepared to see many errors when your corpus is still fresh.1)
POPFile offers users an opportunity to help improve over time by sharing statistics anonymously. To turn this feature on, go to http://127.0.0.1:8080/security
|Messages||Reclassifications||Total Number||Total Number of||Accuracy|
|1 to 1,000||21||1,000||21||97.9%|
|1,001 to 2,000||7||2,000||28||98.6%|
|2,001 to 4,000||8||4,000||36||99.1%|