There is one hard and fast rule for kids and teenagers using the internet: never arrange to meet anyone in the real word after contact online. Unfortunately, this most basic guideline is routinely ignored. According to a study from the cyberspace research unit at the University of Central Lancashire, one in 10 chat users has attended a face-to-face meeting with another chat user.

Even more worrying, of the nearly 1,400 children questioned in the five-year study, five in seven chat users who went to face-to-face meetings knew the rule. As the study argues, scare tactics may actually backfire, as blanket warnings could encourage risk taking from thrill-seeking kids. It says guidelines do not recognise the distinction between risky and non-risky online contexts, nor do they clearly state the logic underpinning the safety guidelines.

"Children who engage in face-to-face meetings often see the process as exciting and adventurous," the study finds. Rachel O'Connell, one of its authors, calls for a more nuanced approach, focusing on education and explanation rather than hectoring. "There has to be a creative blend of solutions for children of different age groups and the rationale has to be explained," says O'Connell.

There has been only a handful of arrests after contact was made on the internet. Two years ago, a 33-year-old man was convicted of unlawful sexual intercourse after raping a 13-year-old girl he met in a chatroom. The paucity of brutal incidents after contact through the internet is no consolation for the victims and their families, and the potential for harm is enormous.

Debate in the UK on the use of chatrooms by children was fuelled this summer by speculation, during the hunt for the missing Soham schoolgirls, Jessica Chapman and Holly Wells, that the girls had made contact with their abductor online. According to the Parents Information Network, an independent service for parents, there are now about 5 million children online in the UK. Of those, more than 1 million are under 14.

It is also estimated that chatrooms are popular with 23% of children, but in the 15- to 16-year-old age group, about 41% use online chat. The Home Office set up a taskforce on child protection on the internet last year. It is due to issue its final report this month. A draft report obtained by the Guardian says that most paedophile activity is covered by legislation.

However, a taskforce subgroup is developing proposals to "tackle predatory behaviour before a sex offence has taken place where there is evidence of an intention to abuse."

The proposal, which is bound to spark alarm among civil liberties groups, will apply offline as well as online. Separately, the framework of the law and penalties for sex offences is also under review. Meanwhile, the draft document has listed best practices for chatroom, instant messaging and for providers of web content.

These include: clear and accessible safety messages designed for parents and children; safety features such as ignore buttons, alert buttons, reporting mechanisms, advice on handling abusive chatters; and easy availability of filtering mechanisms that pick up bad language or prevent children giving out their email addresses.

Other good practices include the prominent display of information on the kind of service offered and the audience at which it is aimed. For example, is the chat moderated or unmoderated? If moderated, what form of moderation is used?

Moderation or supervision of chatrooms can take several forms. Some companies used software to screen out material that identifies users, others use human beings to keep an eye on the content. This can be done by screening messages before they are posted or on a less rigorous basis, with the moderator occasionally dropping into a chatroom to see that everything is OK. For Tom Beaney, head of new business inquiries at Neo1, a company that employs 45 human moderators and provides moderation services for Teletext, Flextech and Handbag.com, unmoderated chatrooms are unsafe and should be closed down.

That is not about to happen because big online brands need chatrooms to boost traffic. "A chatroom is like a swimming pool," Beaney said. "Just as you wouldn't have a swimming pool without lifeguards, so you shouldn't have a chatroom without human moderators." Human moderators may be a good option but it is no panacea.

Companies would have to take extreme care in recruiting, screening, training and monitoring moderators, given their potentially influential position and their access to children. Currently, moderators cannot be screened for criminal records, as their job does not involve physical contact with children.

Ewan MacLeod, the managing director of Neo1, says his company has a rigorous five-week training period for moderators. They are taught what makes a good community, good moderation techniques, what constitutes libel, actionable phrases and are given horror stories.

There are five levels of moderators, with five the highest. Moderators beyond level five become duty managers, directly responsible for the content and performance of each of Neo1's chatrooms and discussion boards. "Being a moderator is not an easy job, it's very intense and we don't want somebody who is in it just for the money," says MacLeod.

"We want people who are passionate and we want to get across that they are holding people's reputations - not just mine but also our clients' - in their hands. It is about three to six months before we let some body loose in a live environment." MacLeod acknowledges that not all chatrooms can be moderated - it takes only two people to set up their own chatroom - but he thinks the big brands such as AOL and Freeserve have to have human moderators to protect their communities and themselves.

O'Connell believes that human moderators are one way to go. But she puts her faith in educational programmes as the best way of tackling the issue of children and internet safety. "It is imperative for educationalists and child welfare experts to recognise the complexities inherent in online transactions," she says, "and to develop educational programmes that cater for the range of situations children may find themselves facing.

"The effectiveness of the current safety guidelines is greatly restricted because they are so limited in terms of relevance and applicability across a range of online contexts."