[HardyLeung]

Nice. One improvement would be to throw in letters in frequencies proportional to the likelihood of appearance in actual text. For example:

http://www.math.cornell.edu/~mec/2003-2004/cryptography/subs...

Though this is a simple problem, if you are picky there are actually other considerations... how well do the words intersect (do you prefer more of a sparse tree-like structure, or maximal mesh-like intersection). What is the optimal dimension of a puzzle given the input. How the order of words (random vs. longest word first) affect the quality of the puzzle. etc. I have a half-done word search puzzle app (with a twist) so this piques my interest.

But I like this one nice and clean.

[Robadob]

I'd prefer the challenge of seeing how many sub-strings from the actual words you could embed in the word-search, without having duplicates of the full strings.

Bit of misdirection for a harder challenge.