Context :

Aims : We present an empirical working model for sunspot umbrae which equally describes observed continuum intensities and line profiles .

Methods : The wings of the infrared Ca II lines depend sensitively on the temperature gradient at -0.6 < log \tau _ { 0.5 } < +0.3 but not essentially on the absolute value of T .
These lines are observed to remain almost unchanged from photosphere to umbra and are thus insensitive to parasitic light .
It is also shown that the infrared K I 7699Å line is suitable for umbral spectroscopy since it is not seriously blended , its continuum is well defined and it is less influenced by parasitic light as compared to lines in the visible spectrum , due to the smaller umbal contrast .

Results : Calculations show that the umbral gradient dT / d \tau , required to fit the Ca II triplet lines , strongly conflicts with the observed profiles of K I 7699 , Na D _ { 2 } and Fe I 5434 ( g=0 ) , even when assuming vanishing Fe II lines for a maximum correction of parasitic light .
It is shown that the discrepancy from the different line profiles may be removed by adopting an opacity enhancement as introduced by Zwaan ( 1974 ) from a discussion of continuum contrasts alone .
The finally proposed umbral working model is very close to a scaled model of the quiet Sun with T _ { eff } = 4000 K thus resembling a M0 rather than a K5 stellar atmosphere .

Conclusions :