Our public library system will be choosing a new ILS ("Integrated Library System" -- geekspeak for "card catalog") to replace Sirsi-Dynix Horizon. This will serve a 30-odd member public library system in Wisconsin.
We have four systems that have been vetted by the tech guy and the catalog goddess, and a committee will be talking with sales reps to make the choice. I'm on the committee, even though my library serves a village of 521 people.
I'd like to hear from others who have made a similar move -- horror stories and success stories.

Haven't done it - yet. I'm not really migrating - just going to upload a bunch of records purchased from our book vendor that I've been sitting on, waiting for my turn to be set up. I don't have a choice in the system I'm getting.

Are you considering an open source system?

We also have Horizon at our Academic Medical library. We considered Symphony and looked at an open source. Can't remember the name. We have to make a decision soon. OCLC is working on something. Should be interesting.

We are considering one open source system (Koha) which hasn't been used much by consortia like ours. We're also considering the "big boys" like Symphony and two others I can't recall offhand. We're fortunate to have some good computer whizzes (including one who was never tainted by Library School) who might be able to make open source work.
Me -- I'm happy with LibraryThing.

we're part of a consortium which uses Symphony (just upgraded from the last version of SIRSI Java). It's fine but it's been so long since I used anything else that I have nothing to compare it to. I've been through a couple of migrations (home-grown to DRA, DRA to SIRSI and then various SIRSI upgrades. Be really careful in mapping your data and expect to have a lot of clean-up. The home-grown (non-MARC) to DRA migration was the worst. After that it was much easier since all were MARC-based.

I'm on Koha currently. They've said that Koha will support true consortia in October with the release of 3.2. Currently, they just have IndependentBranches setting for different libraries.

About a year ago my library (also a small system--our town has a population of somewhere around 5,000) migrated to an open-source ILS called Evergreen based out of Georgia. I'm definitely no expert on this but our cataloger and tech services librarian gives really good reviews for it. And as one who is required to use the system on a daily basis, I don't really have many complaints. It's fairly easy to navigate and understand. Definitely better than our previous ILS (which was also Horizon). I would definitely recommend Evergreen :)

I'm in the same boat as you. The library where I work is currently using Horizon, and we're right in the middle of choosing a new system. I was chosen to be in a focus group, giving my input on what we like and what we don't like about Horizon, as well as our "wish list" of things we want our new catalog to do. I just can't wait to see what our new catalog will look like. Horizon is kind of disappointing.

I'm the cataloguer (in fact the entire tech svcs dept) for a college library in a public-college consortium. We were on Dynix, then Horizon, soon Symphony/Unicorn. So far, I have experienced Symphony only in a test installation, and during setup, but approaching it as an experienced Horizon user, I find that using Symphony is like stepping back twenty years towards something far more cumbersome, inefficient, and marred by many ill-considered, even bizarre, design choices.

Among the latter I would include the three-level architecture (bib record-call# record-item record), the requirement that every item have a call number (and the creation of system generated random-number call numbers for those that don't--eg our entire fiction collection), the use of pseudo-borrowers (e.g. a borrower named 'mending' or 'bindery') in place of status codes and the exclusive use of circulation procedures for simple status changes, the suppression of collection codes ("home locations") in online displays (in favor of the current status), the upper-casing of all call numbers, the complete dependence on 'wizards' for all tasks (you cannot look at a record and then decide to edit it--you have to close the 'search' wizard and start all over again in the 'modify title wizard'), the incredibly convoluted process for exporting records, the lack of many of Horizon's options (such as the ability to choose whether to overlay records or not at the time of import) and Horizon's features (such as scheduled alternative (vacation) addresses for patrons and pop-up checkin/checkout notes), the treatment of Unicode in records as an extra-cost option, and the complete lack of batch-processing/editing facilities.

So far ours is not a horror story, but I fear it may end up that way. Good luck.