Edward seemed to do a good job improving your code with your method of finding primes by checking divisibility by repeatedly dividing.  

However, if your task is to quickly generate a list of primes in a large range, using a prime number sieve (e.g., [Sieve of Eratosthenes](http://en.wikipedia.org/wiki/Sieve_of_Eratosthenes) or an optimized version like Sieve of Atkin or wheel sieves) is a better method.

Your method (as well as Edward's improvement by reducing redundant factors) takes \$O(N^{1.5})\$ to generate a list of primes from 1 to \$N\$ (\$\Sigma_{k=1}^{N} \sqrt{k} \sim O(N^{1.5}) \$).  The Sieve of Eratosthenes takes \$O(N)\$ time to generate the primes from 1 to \$N\$ and the optimized variations are asymptotically faster and can take \$O(N/\log \log N)\$, though you should note for \$N = 2^{32} \sim 4 \;{\rm billion}\$ (common maximum value for an unsigned int), \$\log \log N = 5\$ and for \$N = 2^{64} \sim 10^{19}\$(common maximum value of an unsigned long long int), \$\log \log N = 6\$).

If you need to find larger primes (for example doing RSA and requiring 512-bit primes) or primes in a very small range, its probably better by first checking divisibility against a list of small primes (first 1000 primes or so) and then switch to a probabilistic primality test like [Miller-Rabin](http://en.wikipedia.org/wiki/Miller-Rabin_primality_test) or [Baillie-PSW](http://en.wikipedia.org/wiki/Baillie-PSW_primality_test).