I'm guessing that I am not the only person who tends to draw an exaggerated amount of inferences about strangers from looking at their faces. For example, I might see a person getting onto the bus and think, "hmmm, that person looks kind", or mischievous, or even intelligent. Of course, I am fully aware that I am basing these inferences on nothing by a fleeting first look, and I've had my fair deal of experiences in which I've been proven wrong, but really this has never stopped me from doing this. It's fun.

Obviously, the human face is an incredibly rich source of interpersonal information when it comes to assessing a person's mood. In these cases, we can rely on fairly transient facial expressions to give us a snapshot impression about a person's inner state, but when it comes to using more stable facial features in order to assess more general character traits one should advise caution. For one, we are certainly less accurate (e.g. people judge people's extroversion and conscientiousness at a rate that is only a little above chance), and some psychologists might argue that even when judging stable characteristics it is the transient expressions we are concentrating on.

However, there are certain stable facial features which are closely related to adolescent hormone balances, and thus somewhat linked to behavioral tendencies. In such cases, there might be due reason to assume that people have learned to intuitively associate certain traits with these particular facial features. One such example seems to be the width of male faces.
Facial width it turns out is associated with adolescent testosterone levels and has been experimentally linked to

"other-rated dominance, reactive aggression [...] and [even] the number of penalties in ice hockey."

Given such a link, a group of psychologists has now raised the question whether humans may use facial width to make inferences about character traits such as "trustworthiness"; or more precisely, they asked whether greater facial width may be linked to reduced perceived trustworthiness. In a series of experiments these psychologists, whose findings appear in the journal Psychological Science, took special care to investigate whether people at all exhibit behavioral patterns that relate to facial width as a predictor of trustworthiness, whether people with wider faces indeed behaved in a less trustworthy manner, and whether adjustments of facial width (using computer visual effects) could be used to manipulate people's perceived trustworthiness of strangers.

The experiments were as follows:
For one set of experiments, participants interacted in a so called "trust game", which is a widely used game in economics research. In the trust game, one participant is endowed with a certain amount of money (let's say 5$). The participant get's to decide whether to end the game and keep the money, or whether to "invest" it. Investment in this game means handing over the money to another participant, in whose hands the original investment is multiplied (let's say doubled). The trust element comes to play in this game after the first participant invests and the money has been multiplied, because now the second player receives full autonomy over how much money to return to the first player. In cases of extreme selfishness (or what classical economists would call "rationality") the second player might keep the entire investment (plus the increased returns) for herself, in which case the original investor is left with nothing.

Essentially then, players in this game must decide whether or not they trust the other participant in the game to return an amount that is greater than their original investment. Otherwise they will be better off (or at least they will have earned more money) by not investing at all.
When playing this game in the Stirrat and Perret experiments, participants received no other information about the person they were playing with than a picture of how their counterpart supposedly looked like (in fact people were shown randomized pictures of peoples faces with their hair and clothes obscured).

After running this experiment for 143 students and analyzing the data, the researchers found that

"Women, on average, trusted 45% and reciprocated with 69% of their counterparts. The facial-width  ratios of female participants did not correlate with any of their trust game behavior."

For men, in contrast, trusting ran at around 51% and reciprocation occurred with 72%, on average.

"The facial-width ratios of male participants showed no relation to trust decisions but did relate to decisions to reciprocate. Male participants with higher facial-width ratios (wide faces) were more likely to exploit their counterparts' trust than were male participants with lower facial width."

In other words,

"the ratio of facial width to height predicts male reciprocation behavior in trust games such that wider faced males are more likely to exploit trust than are slimmer faced males".

Further statistical analysis revealed that, besides behaving in a less trustworthy manner, men with wider faces were also trusted less; that is they were intuitively judged as less trustworthy.

Adding to these findings, the researchers then used their photoshop skills to construct fake images of faces which were either artificially widened or slimmed down, and again looked at people's perceived trustworthiness for the people portrayed on these pictures. Below you find one example of such a wide-narrow pair.

The study showed that

"Of the participants, 120 showed no bias or more often chose the images with higher facial-width ratio as more trustworthy, and 165 chose the images with lower facial-width ratio as more trustworthy. Participants were significantly more likely to choose the images with lower facial-width ratio as trustworthy."

The result was especially pronounced for female judges who scored low on self-rated dominance. This last finding is especially interesting from an evolutionary perspective, where particular sensitivity to clues regarding trustworthiness might prove more adaptive the more vulnerable the individual. This would also speak to related claims that facial dimorphism such as the width-height ratio used in this study may be due to sexual selection.
Whatever the reasons for these findings, the results should lead to a number of interesting follow-up studies and conversations.
The actual study raises some additional interesting points in its general conclusions section which I'm going to leave for you to read-up on yourself, as I'd have to write at least another page to cover this in sufficient detail...so is life.

Main Reference:

Stirrat, M. (2010-03-01) Valid Facial Cues to Cooperation and Trust: Male Facial Width and Trustworthiness. Psychological Science, 21(3), 349-354. DOI: 10.1177/0956797610362647