Incident meter: (one shadow reading + one highlight reading) / 2.0

If no shadow, simulate one. Nothing can go wrong.

You can also read markbarendt's primer on incident metering.