All of them.
I never look over the MM stats for normal animals without thinking: "The people who wrote this have never interacted with a real animal." Dex, Int, and Wis are almost always wrong (look at the domestic cat stats, then follow a cat around for a day observing his behavior), and natural attack modes are neglected more often than not. All equines and ungulates have a rear kick attack and it packs a wallop, I'd say a 3x crit, not to mention they can do damage just by stepping or rolling over on you. A friend of mine once used a stable where a man was found dead in a stall, pants around his ankles, hoofprint on his forehead, and the normally-docile mare he'd tried to violate standing quietly. (A constant parade of women went past the stall that day, bearing treats for her.) You know the best thing to do if you meet an angry bear? Climb a tree more than 35 feet high. Lower, and black bears can follow you while grizzlies can just shake you down by whipping the tree back and forth. Camelids have spit attacks, speed, the rear kick, and stamina like you wouldn't believe. Donkeys are smart. Ranchers who incorporate llamas or donkeys into their sheep herds have much less of a problem with predators - including wolves - than ranchers who do not. As for wolves and other canids, their pack tactics and intelligence are not reflected in their MM write-ups. Proboscideans and cetaceans are highly intelligent and can communicate at great distances via infrasound. Turkey vultures have a projectivel vomit attack which is every bit as good as a skunk's spray. Cats have bacteria on their teeth which, buried deep in the puncture wound, settles in and infects. I couldn't use my hand for three days after my cat bit me when I accidentally grabbed his tail stump when we were changing the bandage after his amputation; even lowering it caused excruciating pain, and it swelled up like a basketball.
The dietary strategies and diurnal cycles of animals are never referred to in the MM. Did you know that many sharks, like many whales, are vegetarians or rely on large amounts of small prey? Or that even predatory sharks, like the bull and the great white, don't try to eat humans? Attacks agains humans are a result of a strike against movement in the water similar to suitable prey. The shark grabs, tests the prey with his teeth, and then spits it out if it's too bony, which most humans are. Of course the testing process usually rips the body up so much it can't survive, and sometimes the shark will swallow a bit. Black bears, on the other hand, will deliberately prey on humans if they find them isolated, and they can do so with considerable stealth. Brown bears, including grizzlies, are unlikely to attack humans unless they feel threatened or unless they have eaten enough garbage to associate the smell of humans with food - which is the real reason the park rangers are so fussy about you cleaning up your campsite.
And dire animals, my God, all they ever think of to do with prehistoric versions of animals is make them bigger and give them more absurd sets of teeth. The short-faced bear was bigger than a grizzly and fast enough to run down prairie animals, whose turns of speed are astonishing. The dire wolf is nothing like the MM worg, as it filled the ecological niche of the hyena - slower and stronger than the timber wolf, it had enormous teeth evolved to crush the massive bones of prey like the Western Camel (10 feet tall). Ungulates with long horns, like American bison antiquus and the European aurochs, used them for defense of their herds in combination with tremendous strength, weight, and agility. Mammoths (including the Imperial mammoth, not wooly but 18 feet high at the shoulder) had enormous curving tusks useful for scraping away snow, digging out water holes, and tossing predators aside. All probiscideans should get a slam attack from the trunk, a stomp attack from the feet, and an appropriate tusk attack, as well as the capacity to hurl a predator several feet and perform appropriate fine manipulations with the delicate fingers at the ends of their trunks.
People don't know what to do with sabertooths, which is not surprising given that no modern animals have this adaptation. It appears to be most adaptive in environments with lots of large prey species with plenty of flesh to grip. The sabertooth mouth is arranged with the two prominent fangs off to the side and a jaw that curves in front of that, with a set of incisors that can grip or rip without interference from the larger teeth. Slashing the jugular and then gripping the prey while it bleeds out seems probable. Like sharks, they are likely to scorn bony prey like humans, which doesn't mean they couldn't kill us really good, just that they wouldn't actively prey on us.
When I use real animals, I use my personal experience and my reference books. Our group is experimenting with rules for advancing animals that gets them more skills and more appropriate feats. Although a decent-sized party is unlikely to come into direct conflict with wild animals, because they make so much noise and the animals are out to survive and know enough to evade danger, a properly-run predator, prey, or companion animal should be a significant challenge to an individual PC and can provide welcome texture and variety to a campaign. Stealth attacks by predators too old or wounded to hunt normally can be devastating to agricultural communities or campers (cf, Night of the Grizzlies).
And, as we all know, that rabbit's dynamite.
