I became interested in studying first impressions after my research group discovered that such impressions can predict the outcomes of important political elections. First impressions mattered. I was in my first few years as an assistant professor at Princeton with a tiny lab consisting of one half-time research assistant and two graduate students. There was no easy way to collect data online at the time, so we participated in one of the 'questionnaire days' organized by the department of psychology. These Q days were advertised among students on campus, and those willing to trade an hour of their time for $10 or so were handed a thick bundle of different questionnaires to fill out. Buried among those were some of our questionnaires presenting pairs of images of the winner and the runner-up from all the United States Senate races for 2000 and 2002, excluding races with highly recognizable politicians such as Hillary Clinton and John Kerry. Different students were assigned to different questions, for example 'Who looks more competent?' and 'Who looks more honest?' We were hoping that some of these questions would predict who won the elections. When we analyzed the data, our hopes were surpassed. Judgments of who appeared more competent predicted about 70 per cent of the elections. A general rule of science is that results should be replicated, especially if these results are surprising. So we put everything else on hold and started preparing new questionnaires. We ultimately replicated our initial results and eventually wrote up the findings. The paper was published in Science. This spurred a number of replications by different research groups and in different countries. This was not just a US election phenomenon. Unexpected findings are typically met with skepticism. Before our findings were published, I applied for funding for this research. The way funding applications work is that you write a research proposal and that proposal is reviewed by anonymous reviewers. The reviews covered the whole range, from the very positive to the very negative. One of the reviewers basically wrote that the kinds of effects we had documented - naive judgments from facial appearance predicting political elections - must occur only in my lab. In their words: 'Before I would find these proposed studies at all compelling, I would like to see some evidence that this situation occurs anywhere outside of the PI's laboratory.' The PI stands for me, the principal investigator. Needless to say, I did not get the funding. After the findings were published, I received my first hate email. I still don't understand what instigated it, but its author was pissed off with our 'trivial' results. Buried among the profanities, he actually had a plausible alternative explanation of our findings. According to him, it was patently obvious that the observed effects were due to media exposure. Although our participants did not explicitly recognize the faces of the politicians, they must have been exposed to these faces before, and this exposure made them rate more familiar politicians as more competent. If more familiar politicians are more likely to be the winners, this could explain our results. Though plausible, this hypothesis turned out to be false. The right way to question unexpected findings is to conduct replication studies and test alternative explanations. Political scientists were the first to test trivial explanations, such as differences in the pictures' image quality or in campaign spending. But such differences could not explain the appearance effects on election outcomes, nor could differences in gender and race. In fact, we obtained our best results when the prediction was limited to elections in which the candidates were matched on race and gender. Familiarity with the candidates' faces could not explain the effects either. My favorite replication was a study conducted by John Antonakis and Olaf Dalgas of the University of Lausanne, in which Swiss children's judgments predicted the outcomes of French parliamentary elections. Some of the kids in the study were not even born when these elections were conducted. In other replications in Europe, judgments of US and Swedish participants predicted the outcomes of Finnish elections. With one of my students and a Bulgarian colleague, we used judgments of Americans to predict the 2011 presidential elections in Bulgaria, the country where I grew up. These elections are interesting because there is a very low threshold for entering the race and, as a result, there are many candidates: 18 in this particular year. It was sad to see one of my professors from Sofia University doing very poorly in terms of both appearance judgments and electoral votes. What makes you a good professor does not necessarily make you a good politician. In a final twist, Gabriel Lenz and Chappell Lawson, political scientists at Berkeley and MIT respectively, had American and Indian participants rate the faces of Mexican and Brazilian politicians. Although the cultures of the raters and the politicians were deliberately chosen to be very different, the raters agreed in their judgments of the politicians, and these judgments predicted the outcomes of the elections.