The U.S. Department of Labor claimed last week that Google systemically pays its female employees less than their male counterparts. The tech giant immediately said it was “taken aback” by the allegation, especially since it had announced on Twitter just days before that it had equalized pay across race and gender globally.

Let’s make every day #EqualPayDay. All employers can take steps to eliminate the gender and race pay gaps, today → https://t.co/KTuGTJMV16pic.twitter.com/rBciSK21uF

— Google (@Google) April 4, 2017

To defend itself, Google wrote a blog post explaining how it uses a “gender-blind” approach to determine compensation. According to Google, employee salary is calculated by analysts who have no information about an individual’s gender. Instead, they rely on factors like role, job level, location of the job, and current and recent performance ratings. “Google conducts rigorous, annual analyses so that our pay practices remain aligned with our commitment to equal pay practices,” Eileen Naughton, vice president of people operations, said in the blog post.

Gender-blind compensation processes stem from the success of gender-blind hiring, which many companies are opting for to combat sex and gender-based prejudice. “The process removes knowledge of gender for as long as possible to allow for unbiased decisions triggered by unconscious bias,” said Azmat Mohammed, director at Britain’s Institute of Recruiters. One experiment from tech recruiter Speak With a Geek found that just 5 percent of women were selected for interviews when employers were given complete profiles of applicants. But when identifying details were removed, that number jumped to 54 percent for the same candidates by the same employers.

However, gender-blind techniques only go so far. “The reality is, it is only good in regards to creating a shortlist to interview, the actual hiring is still a flawed process in terms of unconscious bias,” said Mohammed.

It's even tougher for current employees, who have more exposure to bias with regard to compensation or promotion.

In Google's case, the allegation of uneven compensation came to light after the Department of Labor sued the company in January in an effort to get employee compensation data surrounding a compliance audit. In a federal court hearing for the lawsuit, the agency said it uncovered “systemic compensation disparities against women pretty much across the entire workforce.” The Department of Labor notes that the investigation is incomplete but there is “compelling evidence” of “very significant discrimination against women.”

Google immediately denied the accusation, claiming that its “comprehensive and robust analysis of pay across genders” found no sign of a pay gap and accusing the government of lacking proof.

As strict as Google's methods may be, there is still an unavoidable human element that introduces the possibility of bias, whether it be conscious or unconscious. While analysts don’t know gender, managers do and they have the final say about compensation. Google says compensation can be adjusted based on a manager’s recommendation if “they cite a legitimate adjustment rational.”

“Because there is some leeway in what managers can do in terms of compensation and bonus, they may unevenly negotiate for more compensation for men than women,” says Elizabeth Haines, a professor of psychology at William Paterson University who studies unconscious bias and sexism. “Managers may unconsciously believe that If they don’t provide extra compensation to male employees, they will more likely to look around.” Google also has not addressed how compensation gets calculated when employees get competing job offers or negotiate outside the annual cycle.

Haines points out there’s a second area where gender prejudice could filter in: the people conducting performance ratings. Google says “current and recent performance ratings” are on of the several variables used by analysts to come up with compensation so if those evaluations contain bias, they can manifest in their pay equity analysis.

There is still an unavoidable human element that introduces the possibility of bias

“Subjective perceptions can bias the objective performance ratings,” said Haines. “When women act assertively, with power, and behave independently they are routinely disliked, rejected and even sabotaged by their coworkers because they are seen as lacking in warmth and niceness. And women are held to very high standards for being nice, communal, kind, and concerned with others’ goals over their own. Men are not held to these same standards.”

Google’s most recent internal analysis took place in the tail end of the 2016 calendar year across 52 job categories and no gap was found, Naughton wrote in the blog post. When a statistically significant difference is found in compensation adjustments, the company modifies salary accordingly. This methodology, a step in the right direction, is one of several measures Google has implemented to close the gender pay gap — the company has said it avoids anchoring an employee’s salary offer to their current salary and it makes its auditing process for pay public for other companies to learn from.

When all's said and done, there’s a disconnect between the Department of Labor’s declaration and Google’s claims. On one hand, Google’s analysis is an internal one and not from a neutral third party. On the other hand, the Department of Labor isn’t exactly an impartial voice either. The accusation was made amidst a lawsuit hearing where they used the pay gap claim to argue that Google should release more data on pre-2015 salary — a request Google calls a “fishing expedition.”

Recent analysis from the Pew Research Center found that women (both full and part-time employees) earned 83 cents to the man's dollar in 2015, and one study found that gender pay inequality won't be fixed until 2058.

“Can this be done sooner? Of course it can, but not without addressing how human perception – by both men and women – affect the way in which we unconsciously evaluate, judge, and hold women and men to different standards for the same behavior,” said Haines. “Keeping track of the rate at which men and women are adjusted is a first step. Keeping track of the reasons why managers adjust compensation is clearly important. And next, keeping the reasons consistent across gender.”