A previous CD post pointed out one of the problems with historcial median household income from the Census Bureau income data: It doesn't adjust the declining household size over time. After adjusting for household size, real median income is at an all-time high (see charts above).
Robert Samuelson points out three more problems with poverty and income data from the Census Bureau: a) comparing real household income or poverty rates in 2007 to the year 2000 is unfair because 2000 was an artificially high benchmark because of the "tech bubble," b) immigration distorts commonly cited statistics for both poverty and income, and c) Census figures understate income gains by not counting fringe benefits.
Greg Mankiw cites another problem today with Census income data, citing the NY Times, which reported that "The current poverty measure only counts cash as income, and doesn't include government assistance like food stamps, housing subsidies and tax credits. Such aid has been devised to help support the poor, but its impact is not calculated by the current measure."