Measures to reduce the impact of a flu pandemic, such as closing schools, should not necessarily take place at the beginning of an outbreak, according to computer models.
A report in PLoS Computational Biology argues that starting several weeks later could be more effective.
The researchers said this would have a lesser impact on society and the economy.
Experts believe the work could be important for controlling pandemics.
Governments prepare for worst-case scenarios, such as the emergence of a deadly flu outbreak which spreads across the globe.
The UK's pandemic flu plan sets out how the country will deal with such an outbreak.
In the early stages of an outbreak, vaccines are still being developed so measures rely on slowing the spread of a virus, such as closing schools, asking people with flu to stay at home or prescribing antiviral medicines.
When to start?
This study, by researchers at Imperial College London and Utrecht University in the Netherlands, investigated when those measures should be introduced.
It might be thought that the answer is at the start of the outbreak, but the research suggests this is not always the right one.
The problem comes from measures which the government would only want to apply in the short term, because they are expensive and difficult to maintain.
In this case, the mathematical models show that the timing of measures, such as closing schools or restricting public transport, is important.
If they are introduced when the first cases are recorded, it is costly to society and the economy and there is a large second surge if the measures are lifted.
However, when the researchers delayed the introduction of these measures by several weeks, the size of the pandemic and the peak number of cases were similar to the model in which the measures were introduced at the beginning.
This also has the benefit of reducing the second surge while putting off costly measures.
Dr Deirdre Hollingsworth, of the Centre for Outbreak Analysis and Modelling at Imperial College London, said: "If you take into account the impact that those policies will have on society, it might be better to hold back at the start.
"It's like pacing yourself for a race: while you can cope with a pandemic you don't want to disrupt public life."
Professor Matt Keeling, who models infectious diseases at the University of Warwick, said: "For ages we've always been saying, hit hard and early, a bit like dealing with compound interest.
"However, if you can only hit for so long, then don't do it at the beginning, target the bit you want to control.
"Like all good computer modelling, the conclusions seem obvious afterwards."
The researchers are clear that their work is not a policy document. However, the Department of Health said it monitored all research in the area and noted the findings with interest.
A spokesperson said: "We committed in the public health white paper to publish a revised strategy for pandemic flu, which will consulted on later in the year."
Dr Nim Pathy from Princeton University said: "Experience of the swine flu pandemic showed just how much uncertainty there can be in these early stages, so this work could have important implications for the practicalities of pandemic control.
"If potentially costly interventions can be delayed even by a matter of weeks, this may buy valuable time to gather important clinical, virological and epidemiological information. Such data would be key in guiding flexible, responsive pandemic control measures."