dataloader vs import and duplicate handling

I thought we'd be able to use dataloader to load data from source system file in a daily automated fashion. However,

apparently dataloader, which uses import somehow manages to ignore/override import capabilities of updating a record if it already exists!

When I run it in insert mode it creates duplicates, rather than overwriting the existing record.  Is the only way around this to write custom code to

use webservices to load the data (and figure out if record already exists first)?