I recently blogged about an article by Joe Nocera in the New York Times Magazine. Suni Reyent at Seeking Alpha just wrote a response.
One of the common criticisms of VaR by Nassim Taleb (and others) is that it incorrectly models the distribution of stock returns (or of whatever asset that VaR is being applied to) as a normal distribution. This would be a serious flaw, since the distribution of returns of many assets are leptokurtotic (i.e. have fatter tails than a normal distribution).
In this piece, Reyent discusses how VaR is applied in practice, and adds that it often adjusts for the non-normality of returns. Having come from a practitioner's background, she's worked with this stuff often, so she knows what she's talking about.
Read the article here.