A new method is presented to compute age estimates from theoretical isochrones using temperature , luminosity and metallicity data for individual stars .
Based on Bayesian probability theory , this method avoids the systematic biases affecting simpler strategies , and provides reliable estimates of the age probability distribution function for late-type dwarfs .
Basic assumptions about the a priori parameter distribution suitable for the solar neighbourhood are combined with the likelihood assigned to the observed data to yield the complete posterior age probability .
This method is especially relevant for G dwarfs in the 3-15 Gyr range of ages , crucial to the study of the chemical and dynamical history of the Galaxy .
In many cases , it yields markedly different results from the traditional approach of reading the derived age from the isochrone nearest to the data point .
We show that the strongest effect affecting the traditional approach is that of strongly favoring computed ages near the end-of-main-sequence lifetime .
The Bayesian method compensates for this potential bias and generally assigns much higher probabilities to lower , main-sequence ages , compared to short-lived evolved stages .
This has a strong influence on any application to galactic studies , especially given the present uncertainties on the absolute temperature scale of the stellar evolution models .
In particular , the known mismatch between the model predictions and the observations for moderately metal-poor dwarfs ( -1 < [ Fe / H ] < -0.3 ) has a dramatic effect on the traditional age determination .

We apply our method to the classic sample of Edvardsson et al .
( 1993 ) , who derived the age-metallicity relation ( AMR ) of a sample of 189 field dwarfs with precisely determined abundances .
We show how most of the observed scatter in the AMR is caused by the interplay between the systematic biases affecting the traditional age determination , the colour mismatch with the evolution models , and the presence of undetected binaries .
Using new parallax , temperature and metallicity data , our age determination for the Edvardsson et al .
sample indicates that the intrinsic dispersion in the AMR is at most 0.15 dex and probably lower .
In particular , we show that old , metal-rich objects ( [ Fe / H ] \sim 0.0 dex , age > 5 
Gyr ) and young , metal-poor objects ( [ Fe / H ] < -0.5 dex , age < 6 
Gyr ) in many observed AMR plots are artifacts caused by a too simple treatment of the age determination , and that the Galactic AMR is monotonically increasing and rather well-defined .
Incidentally , our results tend to restore confidence in the method of age determination from chromospheric activity for field dwarfs .