How would I work out a distribution free 95% confidence interval of the median if given a large sample?

To get the 95% CI I use \mbox{estimate} \pm 1.96 \times \mbox{standard error}. I can work out the estimate of the median, which is just the sample median.

However, to find the standard error do I use var(m) \approx \frac{1}{4nf(m)^2} and then square root that or would that not be distribution free??

Would I use the normal approximation to the binomial? Bi(n,0.5) \approx^d N(0.5n,0.25n)?

This page describes a general straightforward procedure for finding a 95% confidence interval around any quantile, and uses the median (50th) as the example.

This is for a large sample, however.

http://www-users.york.ac.uk/~mb55/intro/cicent.htm (http://www-users.york.ac.uk/%7Emb55/intro/cicent.htm)