Philadelphia city commissioners are investigating an unusual series of over-votes in last year's primary election - 83 voting divisions citywide where the official vote totals were bigger than the recorded number of voters who showed up.

In most locations, the discrepancies were small, just a handful of votes. In many instances, minor procedural mistakes could account for the anomalies.

But so far, the bulk of the over-voting has not been explained.

Until they understand what happened, the commissioners say, they cannot rule out the possibility of deliberate, illegal efforts to run up votes for favored candidates, with the perpetrators losing count as they tried to cover their tracks.

In a situation like that, the tiny numbers of over-votes might be red flags for a much larger problem with the underlying vote totals.

"That's within the realm of possibility, absolutely," said Stephanie Singer, who became chairwoman of the city commissioners in January.

It's also possible the discrepancies were isolated cases with innocent explanations, she said. "We won't know until we know," she said. "When we see things that look very suspicious, then we will go after them."

The biggest mismatch turned out to have a simple cause.

At a 20th Ward polling place near Temple University in North Philadelphia, only six people signed the poll book, required before they were ushered to a voting machine. Yet the official election results had the division providing 80 votes for the various candidates for mayor.

When the commissioners took a closer look, they discovered two divisions were voting in the same place. When poll workers arranged the machines for use on Election Day, a machine set up for one division was actually used in the other division.

The mistake skewed the division-by-division results, but there was no apparent impact on the total votes any candidate received.

"We know there's a problem. We don't know yet what the problem is," said Al Schmidt, who also took office in January as the sole Republican among the three commissioners.

"You have some divisions where the poll books say 100 people showed up to vote and there are 108 votes on the machine. That's a problem," Schmidt said. "In some cases, it looks like it might be a matter of Republicans participating in the Democratic primary. . . . With elections that are so close, like the Republican primary for mayor, it could have affected the outcome of that election."

The over-vote problem was first brought to the commissioners' attention by a couple of Republican Party activists - Adam Lang, a computer network engineer who has been working with voter data for several years, and Joseph DeFelice, an organizer paid by the Republican State Committee to work on the party structure in Philadelphia.

Lang obtained data in January from the new city commissioners showing which voters had voted in which polling places in the May 2011 municipal primary.

Then he compared the numbers to actual vote totals in each division. Typically, the vote totals in the highest-profile races run slightly below the total count of voters in either party, because some voters skip over ballot choices, are inattentive or have no preference among candidates.

But in 83 divisions, about 5 percent of the 1,687 divisions citywide, Lang found vote totals that exceeded the number of qualified voters - 40 divisions with an over-vote for Democratic candidates and 43 for Republican candidates.

DeFelice told the commissioners some of the polling places with over-votes were in the same wards where Republican poll watchers had perennial disputes with the five-member election boards responsible for running the election.

DeFelice was particularly skeptical of results in the Seventh and 19th Wards, where the City Council contest between Maria Quiñones-Sánchez and Danny Savage recorded more than twice as many Democratic votes as in the Democratic mayoral contest between Michael Nutter and Milton Street.

Singer promised DeFelice the commissioners would look into the situation. She and Schmidt have been going through poll books and voting machine results, division by division, trying to figure out what happened.

"It's more than just counting up signatures," Schmidt said. "A lot of the signatures [in voter registration records] don't match at all the signatures in that poll book."

Schmidt said some signatures did not "even remotely" resemble signatures on file. "I don't need to be a handwriting expert to know that something's wrong," he said.

If necessary, Singer said, the commissioners will begin interviewing election board members as part of their probe. Apart from the biggest over-vote, in the North Philadelphia division near Temple, they have no answers so far.

The public official in charge of last year's elections was Singer's predecessor, Margaret "Marge" Tartaglione, whom Singer defeated in the May primary. Tartaglione declined to comment Monday.

Bill Rubin, who was supervisor of elections until he resigned last year to run for Council, said minor discrepancies involving voter counts were not uncommon.

"If it was a huge number or a big disparity, our people would pick it up," he said. "There are a lot of reasons it could happen without there being any fraud or a major conspiracy. Maybe they set up the machine for a Democratic voter when it should be set up for a Republican. Maybe a husband and wife go in, but only one signs the poll book. The poll books get scanned with a wand to record who voted - maybe the wand didn't work and it didn't pick up all the voters."

Frederick Voigt, now the commissioners' legal counsel and former director of the Committee of Seventy, said similar over-votes in the past had generally been traced to two or more divisions' being located in the same place.

"There are mistakes in every election," Voigt said. "You have too many human beings for there not to be mistakes made."

He said the electronic voting machines now in use had eliminated what used to be the most common source of vote-count errors - mistakes made by humans in transcribing the results from voting machines or in entering data inaccurately.