Atul Gawande cannot forget the first time he sliced someone open. "I was in my final year at medical school," he says, "and the surgeon in the operating room just handed me the knife on the spur of the moment and said, 'You start.' The knife was still warm from the autoclave [sterilising machine] and I started cutting down the line he had drawn for the hernia incision. The skin was much tougher and more rubbery than I expected and my first cut barely went through, so I had to make a second. It wasn't frightening or nauseating in any way, but I did have a surreal sense that I was committing a violation. It felt an odd claim to make, that cutting someone open could make them better."

More than 10 years and roughly 4,000 operations later, Gawande has never lost sight of surgery's narrow divide between the sacred and the profane. Even now, before every procedure - no matter how routine - he runs through a checklist of the top three things that could go wrong and mentally prepares for these eventualities. And yet still things go wrong. "Each year I do about 400 operations and have about a 97% success rate," he says. "This means that 12 of my patients - two of whom will probably be dead - leave the theatre worse off than when they came in.

"Most of the time, they are patients whom I think I've done all I possibly could for, so I can accept it. But there are always the others. Six months ago, I was performing a thyroid operation on a teacher my age and damaged a nerve that led to her vocal cords. This woman can no longer talk and has had to give up work. The only way I can live with myself is to try to do all I can for her and understand what I did wrong and do better next time. Being sorry is not enough."

Gawande is a general surgeon at the Brigham and Women's Hospital in Boston and assistant professor at Harvard Medical School. And he still makes mistakes. It's this uncomfortable wound that he has opened up, first in Complications - his bestselling book that was shortlisted for the National Book prize in the US and became the inspiration for the TV series Grey's Anatomy - and now in his follow-up, Better.

"What I'm interested in is failure," he says, "as it's the one area of medicine with which the professionals are often reluctant to engage because the stakes we are playing for are so high. We can fail by putting a decimal point in the wrong place and by not asking the right questions. If you ask any doctor when he or she last made a misdiagnosis, the truthful answer would always be in the last month. We get things wrong and we try to put them right. And, of course, we can fail with a slip of the hand. I once performed an emergency trachaeotomy in which I did everything wrong. I had the wrong knife, the wrong lights and I made the wrong incision. There was blood everywhere and the patient would have died if a colleague hadn't stepped in to help. It was horrific."

If his call for willingness to ruthlessly examine what goes wrong and why is not always shared by other doctors, then neither will be the second part of his prescription for demystifying failure. Mention league tables to most doctors and they start spitting blood, but Gawande is adamant that there is a value in open accountability. "You have to be careful you are comparing like with like," he says, by way of a caveat, "because otherwise the data can be meaningless. For instance, death rates for cardiac surgeons can vary hugely, depending on the age of their patients and the difficulty of the procedures they are undertaking. But we shouldn't allow this to disguise the fact that some surgeons and some hospitals get better results than others. And unless we are prepared to admit this, we'll never improve the overall success rates."

As an example, Gawande cites both the case of the paediatric team at a hospital in Cleveland, Ohio, who virtually single-handedly raised the average life expectancy of cystic fibrosis patients from three years old in 1957 to 33 in 2003, simply by providing aggressive preventive treatment before sufferers became visibly sick from it. You could argue that such successes are largely a matter of money, but Gawande disagrees. "We spend huge amounts of cash on healthcare in the US," he says, "and not all of it is very effective. Infection rates in hospitals are proof of this. Some hospitals have spent a small fortune on portable $5,000 hand-washing machines [superbugs are as rife in America as they are in Britain], but their infection rates have remained as high as before.

"The one hospital in the US that has managed to reduce infection to almost zero did so not by spending money but by analysing procedures. They asked doctors and nurses how they used the system and found that the alcohol gel dispensers were in the wrong place, the gowns they needed for the 3pm ward rounds were in the wrong place and that it was better to have a single stethoscope designated for each patient's sole use than for a doctor to carry his own from patient to patient."

Gawande goes on to argue that though we traditionally associate significant improvements in healthcare with the big breakthroughs in science - such as transplant surgery and gene therapies - much the biggest gains are likely to come from the close attention to the detail of failure. Rather than shrugging their shoulders and boasting that a 98% success rate for any given procedure is a stunning success, doctors really ought to be looking at why the procedure does not work in the other 2% of cases.

Here it all gets tricky, though, because the public's attitude towards failure has become far less tolerant over the past two decades. Any hint of failure or incompetence is usually met with a lawsuit for malpractice and US surgeons pay insurance premiums of between $30,000 and $300,000 per year in case of just such an eventuality. In a way, this is just as it should be, because doctors should be held accountable for their actions. But Gawande believes there is a mismatch between the public's perception of what is possible and what actually is. "People don't truly understand the risks," he says. "Back in the 1960s, about one in 30 babies would die in childbirth; now we've got it down to less than one in 500 and people expect their baby to be born alive. When it isn't, they assume the doctor has messed up when he hasn't necessarily. That's why we have seen such an increase in caesarean sections: doctors just don't want to take the chance."

The downside of risk-averse medicine is that we may not have the bottle to see a treatment through. "Back in the 1950s, kidney disease used to be a killer," Gawande says, "and there was a medical team in the US that was pioneering transplant surgery. Their first 30 patients died. Imagine that. You or I might have stopped at 10, thinking it just wasn't going to work. After 20 deaths, even members of the team began to wonder if they were murderers. But then they got it right, Joe Murray won a Nobel prize and we now have an everyday operation that has saved many thousands of lives."

The trick, Gawande insists, is to never lose sight of the fact that surgery is an imperfect science and that there will inevitably be times when you can only "peek and shriek". "You just have to be the type of person who believes that action is generally better than inaction in times of uncertainty," he says. "There are times when you are operating that everything feels just right and it's as though you're conducting a symphony. But even then you can't always trust that feeling.

"I had one patient for whom I was convinced I had done great work by removing an adrenal tumour, but he still hadn't left the intensive therapy unit six months later. Somehow, he caught an infection and just wasn't strong enough to recover. Our best guess is that someone on the operating team didn't wash their hands properly. And, you know what? It could have been me."

Buy Better: A Surgeon's Notes on Performance by Atul Gawande at the Guardian bookshop