I've just read Elizabeth Anderson's critique of the use of cost-benefit analysis in environmental and workplace safety policy. (It was written in 1995, so I'm a little behind.) It appears in chapter 9 of her book Value in Ethics and Economics and is reprinted in the anthology Philosophy, Politics, and Economics.
It's a good read for those who estimate the value of a statistical life of the willingness to pay for improvements in air quality. I've read some other critiques of cost-benefit analysis and of non-market valuation that aren't nearly as compelling as this one is. (Disclosure: Anderson was my professor for an ethics course back in the late 90s when I was an undergraduate philosophy major at Michigan.)
I will argue that these goods [safety and environmental quality] are not properly regarded as mere commodities. By regarding them only as commodity values, cost-benefit analysis fails to consider the proper roles they occupy in public life.
Her critique of the VSL is basically that, even if you believe that the estimated VSL represents a consistent value of what people would pay for mortality risk reductions, it does not have the normative representation as it is commonly used in policy analysis.
Market norms and social relations do not supply an adequate context for people to autonomously express how they value their lives.
She similarly critiques the use of non-market valuation of environmental goods like clean air. While some of her criticisms are unfair I think (she seems to overlook the fact that economists doing non-market valuation do in fact allow for valuations other than use value), there are many compelling issues to think about. She argues that attitudes towards nature can be "intrinsic evaluative attitudes," which are not subject to monetization or trade-offs. People can value environmental goods "in higher ways" than they value commodities, and using standard welfare economics to measure environmental (non-market) valuation will not capture this.
Her proposed alternative to using CBA to assess policies in these domains is to use "democratic institutions of voice," in which individuals are given autonomy through the democratic process rather than only insofar as they are willing to pay for improvements in safety or the environment.
This chapter is well worth reading for any of us who do CBA or non-market valuation. Even if the conclusions are wrong, we should all think more clearly about the justifications behind our analysis.