There's something daunting about taking a skull that's thousands of years old, often smashed or degraded in key places, and working out what its face would have looked like.
There are no photos to compare with to see if you're wrong, and according to a prominent facial anthropologist, there is not a lot of evidence to back up most of the facial reconstructions we see today.
The University of Wollongong's Dr Susan Hayes has built a career on turning skulls into faces and hopes to share a little bit of her knowledge with a Queensland audience as part of National Science Week.
Dr Hayes, who practices a new approach to what she prefers to call facial estimation, has been involved in several high-profile cases involving both research and police work.
She reconstructed the face of so-called Belanglo 'angel', which would eventually turn out to be missing mother Karlie Jade Pearce-Stevenson.
The anatomy and human biology PhD holder also gave a face to the so-called "hobbit" fossils of Indonesia's island of Flores and recently estimated how both missing toddler William Tyrell and Princess Diana would look today.
But Dr Hayes raised concerns about the accuracy of the majority of facial reconstructions found in museums or released by police in an attempt to generate leads for unsolved crimes.
"Nearly every face that's been estimated in the world, particularly the ones in museums, we don't know how they did them," she said.
"They claim to be using the forensic facial reconstruction method and as I said, it has been systematically evaluated and found to be invalid and largely replaced."
Dr Hayes said only two actual reconstructions were published in a journal, and one of those was hers.
In her interactive lecture at the Queensland Museum on Thursday, Dr Hayes will delve into both the "angel" and "hobbit" cases.
Disregarding both the US and European facial reconstruction models she said had been disproved, Dr Hayes instead used statistical averages, backed up by peer-reviewed research, to create her "facial estimations".
"I wouldn't say that my approach is necessarily more accurate. What my approach is It's actually justified with reference to the research literature.
"But they're all based on statistical averages of human variation and nobody's average."
This proved particularly true with the Pearce-Stevenson case, which provided Dr Hayes with a relatively rare chance to assess her efforts for accuracy.
She cross-referenced three images of the slain mother with 64 other images of young women of similar age, with similar head poses to evaluate her work.
Dr Hayes, who is also master of fine arts, found a computer program probably would have found a match but a few inaccuracies around the nose and mouth may have stopped close friends and family recognising the woman.
"The results indicated that metrically, if you're using mathematical software,you probably would have got a hit, definitely, because they grouped nicely but it's possible that familiar face recognition would have been interrupted because we don't recognise our friends and loved ones metrically."