By Jonathan Amos
BBC News science reporter, San Francisco
Scientists are reporting promising results from the project to drill into the famous San Andreas Fault.
The borehole broke through the fault in August
The 1,300km-long "crack" that runs right through central California has been the cause of many big earthquakes.
The San Andreas Fault Observatory at Depth (Safod) project has now sent a borehole into the active part of this quake "machine", more than 3km down.
Instrumentation will not be fully fitted until 2007, but already Safod is providing fascinating and useful data.
"We're seeing very localised zones of intensely deformed rocks," said Bill Ellsworth, a co-principal investigator on the programme.
"These form something we call fault gouge, material that has been torn apart by past movement on the fault. These zones are very narrow; some are a few metres wide, some are just a finger's width."
The US Geological Survey researcher was speaking here at the American Geophysical Union Fall Meeting.
The drilling project is sited in the town of Parkfield, the self-professed "earthquake capital of the world".
Parkfield experiences small to moderate-sized quakes at regular intervals as the Pacific and North American tectonic plates grind against one another along the San Andreas fault.
As well as inserting instruments into the hole, Safod brings rocks up to the surface to be analysed in the laboratory to learn more about the forces that have been working on them.
Project drilled directly into San Andreas quake zone
It will measure changes in rock before, during, after tremors
Data will be compared with surface measurements
"We've never known the real mechanics of earthquakes because we're always studying them from the surface," explained Dr Ellsworth.
"What we want to do is to get inside to measure the physical quantities that change during the earthquake cycle.
"We know that stresses are released but how big are they and how do they accumulate in time? That's what we'll get by going into the heart of the machine."
The key prize, of course, will be to establish if any precursor signals exist that could be used to predict earthquakes reliably and accurately.