Variable, but improving all the time. That's why the earlier time periods have bigger variation and bigger error bars, to reflect the bigger uncertainties. What we have to be absolutely clear on is that these graphs are just summaries of very, very large datasets. Real climate science doesn't rely on these, or on any single dataset. It aims to unify all the datasets: direct temperature measurements, satellite measurements, ice core data, sediment data, tree ring data, sunspot data... as many different records of past conditions as they can find, with careful and rigorous consideration of the uncertainties in each. It's a common denialist trick to pick one dataset, pick some hole in it (real or otherwise) and declare the whole field discredited. Real science doesn't work that way. If you want to disprove AGW you need to come up with an alternative, detailed physical mechanism that explains all the data better than the current theory.
