I'm working through all 60 concepts at the end of Sklansky and Miller's No Limit Hold 'em: Theory and Practice.
Concept No. 10: Sometimes you should go for a check-raise bluff on the river when a bluff bet would be unprofitable.
Sklansky and Miller are technically correct that you should sometimes check-raise as a bluff, but the advice is still bad. It's confusing for them to state this advice without noting the unusual circumstances under which it holds. They say, "sometimes you should..." Well, yes, but rarely. Readers who take Concept 10 at face value are in danger of trying check-raise bluffs indiscriminately, which would probably be a big mistake. My problem with this advice is actually very similar to Concept No. 4: "Sometimes you should bluff to stop a bluff." In both cases, Sklansky and Miller are missing the main point, an idea that I've attributed to Mike Caro (though I can't find it anywhere - maybe I thought of it myself!): If your opponent bluffs either way too much or way too little, be more willing to check to him. Or, more generally: If you've identified any weakness in your opponent's play (such as bluffing too much), try to put him in situations that will allow you to exploit this (for example, by checking and letting him bluff).
In Concept 4, Sklansky and Miller refer to an opponent who folds way too often to a small bet. You should bluff often against this opponent. Here in Concept 10, they use an example where your opponent bets 5/9 of the pot and is bluffing "the majority of the time." This is too often. (When betting 5/9 of the pot, game theory suggests you should be bluffing only 5/14 = 35.7% of the time.) Against an opponent who bluffs too much, checking is commonly the best play. This is the real lesson of the example they give: If you know your opponent bluffs too much, exploit it!
I know Sklansky and Miller are not talking about game theory here, but I just want to acknowledge that theoretically, it is indeed sometimes profitable to check-raise bluff even if your opponent is playing perfectly. For an optimal strategy, these bluffs are needed to balance the times you check-raise with a strong hand. In this theoretical case, the hands that are best to check-raise bluff with are actually not the very worst hands in your range (which are the best hands to try regular bluff bets with). Instead, the best hands to check-raise bluff with are the very best of the hands that you would otherwise fold. For example, if 66 is the worst hand you would call with in a given situation, 55 is actually the best hand to try a check-raise bluff with. In practice, if you want to try a check-raise bluff, the decision should depend a lot more on your read of your opponent than on your own hand, but I always think it's worth considering what game theory has to say about a situation.