Thecontingent valuation (CV) methodology assigns prices to environmental amenities by asking people how much they would be willing to pay in order to preserve or acquire those amenities. If this measurement procedure is valid, then responses should be sensitive to relevant changes in the amenities being judged and insensitive to irrelevant changes. One apparent demonstration of inappropriate insensitivity is theembedding effect: the observation that people are apparently willing to pay the same amount of money for a good as for a minor subset of that good. This study examined the possibility that the source of this effect lies with each of two (potentially treatable) methodological problems: 1) subjects have difficulty using quantitative (dollar) response modes to express their values; and 2) subjects have difficulty absorbing the essential details of the CV scenarios describing those goods. The study found that 1) subjects showed considerable embedding both with a simple paired-comparison response mode and with a more demanding one requiring direct dollar estimates; 2) embedding was much reduced with the simpler response mode; 3) subjects' preferences with the two response modes were usually inconsistent; 4) when asked to describe the CV scenario that they had just heard, subjects often reported key task details inaccurately; and 5) there was less embedding when tasks were reinterpreted in terms of the questions subjects reported having answered (as opposed to what had actually been asked). These results are discussed in terms of the match between the questions that investigators would like to ask and the ones that subjects are capable of answering.