What is a statistically valid sample?

for example, if I have a database of 100,000 answers to poll questions, how many random answers would I need to select for it to be "statistically valid"? (e.g. 1,000? 500?)