I have 600 vehicles in my fleet that I don?t know their engine hours ?
engine hours are important so I know when to do my preventative
maintenance. For the fleet it?s supposed to be done every 2000 hours

There is a meter on each vehicle that I could send someone out to read
but that is costly so I?m looking for a way to guess/calculate an
average number to set them all at in my reminder software which would
approximate on average the proper schedule.

That is, currently the hour meter reading will be somewhere between 0
and 2000 hours across the fleet, ie some will have just done their
maintenance with their hours reset to 0 while others will be just
about ready for theirs with hours close to 1000. I also allow a
tolerance of +/- 200 hours, ie I can do them 200 hours early or 200
hours late

If the hour readings on the vehicles would be statistically spread
anyway, could I calculate the right number to reset all vehicles in my
system so that I would end up doing about the same amount of
maintenance as if I had the actual hour meter readings?

Commenter gave you good advice on
how to deal with the similar problem in the future.

However, you are asking: For my present situation,
 is there some clever algorithm, which will save me the effort.

Answer is 'no'.

That us using one detail you gave us:
 " some will have just done their
maintenance with their hours reset to 0 while others will be just
about ready for theirs with hours close to 1000 .." (or 2000 I suppose).

indicating that vehicles went through many cycles already.

They all start (if you buy whole bunch at time t) with delta-like
(sharp peak) distribution, which after many cycles evolves into
uniform (flat) distribution.

 That's what you have now.  The average of hours is 1000 but that will not
 help you to decide which truck to send for maintenance.

No algoritm can create missing data in this case.

Process is called diffusion. It is shown here in 2D.
http://www.geocities.com/piratord/browni/Difus.html

RFID (Radio frequency id) tags might give you the data automatically
without  having a human reader do the reading.

How many cars can your maintenance folks do each day or week?

600 divided by 10 or 20 or 50?

One driver per vehicle 5 days or 3 drivers 7/24?

Are there set routes or is the fleet for hire like cabs? 

(Mileage predictable or variable?)

Aren't the drivers responsible enough to ask them to report when
1800-2000 hours are being approached?  Even at 7/24 use, 200 hours is
eight days leeway, and that would be with three or four drivers in a
position to report.  If the vehicles are always driven by just one
person, I reckon they cannot drive more than 12 hours/day, giving them
two weeks to report (and then they should have a greater interest in
the proper maintenance of "their" vehicle).
If they report when 1800 hours is reached, you would presumably have
ample time to arrange to have the vehicle in the shop, both the
logistics of getting it there and planning down-time and substitute
vehicle as well as a constant work level in the shop.

I expect that you would want to establish a database that could
include a record of maintenance and calculate average hours on a
calendar for different types of vehicles, thus allowing a rough
control, having the software highlight vehicles that were over the
average number of calendar days for their type.

I agree with RedFox and MyOarin

- you need to develop an accurate way of collecting the data

Could you tell us a little more about your fleet, are these rental
cars, who are the drivers - that sort of stuff.

For example, is the servicing done in-house, or is it sub contracted.

I've a suspicion that you are trying to apply OR theory for changing
things like street lamp bulbs - but they are inexpensive and a vehicle
failure is very expensive.

There should be a simple way of reliably and inexpensively collecting
the data, by leaching on something else in your 'system'.

Look back on your records to get an estimate of how many calendar days
it takes to  get 2000 engine hours. Call this A. Assuming engine hours
are approximately linear with calendar days, since 200 is 10% of 2000,
you should call in your cars after  A+/-.1A many days.