Start with the physics, because it decides everything else
Both fiber types carry light; they differ in how much light, and how messily. Multimode fiber has a wide core (50 or 62.5 microns) that admits many simultaneous light paths — 'modes'. Those paths arrive at slightly different times, smearing the signal as distance grows; this modal dispersion is why multimode reach is measured in hundreds of metres. Single-mode fiber's core is about 9 microns: only one path fits, the signal stays clean, and reach becomes a question of laser power and receiver sensitivity — kilometres, then tens of kilometres.
The optics differ accordingly. Multimode links use cheap LED/VCSEL optics at 850nm. Single-mode uses lasers at 1310nm or 1550nm. The glass itself inverts the price story — single-mode cable is usually cheaper per metre than OM3/OM4 multimode — while single-mode optics cost more per end. At short reach the cheap multimode optics win; at distance, only single-mode is even possible.
The reach numbers, from real datasheets
Take the figures printed in an actual gigabit media converter datasheet (the ONV1110-SLX-F, featured on this site): multimode reach 0–550 metres; single-mode 0–40 km at 1310nm and 0–120 km at 1550nm. Those numbers frame the industry: gigabit multimode tops out around 550m (275–550m depending on fiber grade and standard), while single-mode starts where multimode stops and runs two orders of magnitude further.
Notice what the numbers imply for design margins. A 400-metre multimode gigabit link is at 73% of its envelope before you count patch panels and splices; a 400-metre single-mode link is at 1% of its envelope. Fiber planted for decades should be chosen with that asymmetry in mind.
Cost: the crossover everyone gets backwards
The folk wisdom says 'multimode is the cheap option'. At the optics, yes: an 850nm multimode SFP is the cheapest module in any catalogue. At the cable, no: single-mode fiber cable is a commodity produced in vastly greater volume (it is what carriers run) and is typically cheaper per metre than laser-optimised multimode. Termination and testing costs are comparable. The total for a short in-building link still favours multimode on module price; beyond a few hundred metres, or wherever conduit labour dominates (it always does), single-mode's total is equal or lower — and it never needs replacing when speeds climb.
The upgrade path seals the argument. Multimode's bandwidth-distance product is a hard ceiling: the same OM3 plant that does 10G at 300m cannot do 40G at distance at all. Single-mode plant installed today carries 1G, 10G and 100G optics across its life. Nobody re-trenches a campus to save on two modules.
So which does your link need?
Inside a building, under ~300 metres, with 1G/10G speeds and existing multimode plant: multimode is rational, and its optics are the cheapest line in the budget. Everything else defaults to single-mode: any run leaving the building, any run you cannot easily re-pull, any distance you are even slightly unsure of, and any link whose speed may grow. The industry pattern in ONV's converter and switch catalogue mirrors this: the same converter accepts multimode dual-fiber, single-mode dual-fiber or single-mode single-fiber modules — the hardware is neutral; the fiber decision is yours, and it should be made for the link's whole life, not its first year.
One special case deserves a sentence: single-fiber (WDM) single-mode, which runs both directions on one strand at two wavelengths. It exists for scarce-duct and leased-fiber situations, and its optics are bought as matched A/B pairs. Scarce strands are the only reason to accept that constraint — with two strands available, dual-fiber single-mode is the simplest long-life choice.
A note on installed plant and honesty in specification
Existing buildings complicate the purity of the choice: a riser already full of multimode should be used for what it can carry (in-building gigabit and 10G within its envelope) while new inter-building runs go single-mode — mixed estates are normal and manageable when each link is documented for what it is. The documentation point matters more than it sounds: most fiber grief in maintenance comes not from the wrong fiber but from an unlabelled one, where a technician patches a single-mode link with a multimode module and troubleshoots a 'dead' link that was never going to light.
When you read any fiber datasheet — including the converter datasheets quoted on this site — hold the vendor to the same standard applied throughout this guide: reach stated per wavelength and fiber type, in a table, in the model's own specification. 'Up to 120 km' as a headline with no per-module table is marketing; the table is engineering. The ONV converter datasheet prints multimode 0–550m and single-mode 0–40 km (1310nm) / 0–120 km (1550nm) side by side, which is the format to demand from anyone.
Worked examples: three links, three answers
Link one: a 180-metre riser inside a hospital tower, gigabit today, 10G within five years. Multimode OM3 carries gigabit to 550m and 10G to 300m — this link fits both, today and at the upgrade, on the cheaper optics. Multimode is the rational buy, with one condition: the run is documented as multimode in the building record, because the day someone patches it with single-mode modules it becomes a ghost fault. Link two: the same hospital's link to a detached imaging building, 900 metres of duct. Multimode is not a candidate at all; single-mode at 1310nm covers it at 2% of the optics' reach envelope. The interesting decision is not the fiber but the strand count — pull spare strands; the duct labour is the cost, the glass is nearly free.
Link three: a farm estate with gates and barns at 300m, 1.2 km and 3.5 km, cameras only. Every instinct says 'cameras are slow, use the cheap option' — but the cheap option does not exist at 1.2 km. A single-mode design serves all three distances with one module type, one spare on the shelf and no exceptions to remember. That uniformity is worth more than the small per-link saving a mixed design might find at 300m: estates are maintained by whoever is available, and homogeneous infrastructure forgives its maintainers. The principle generalises: choose per-link only when the estate is small and static; choose per-estate when it will grow or be maintained by strangers.
The decision, compressed
If you remember nothing else: multimode is a room-and-riser technology with a hard distance ceiling and cheap optics; single-mode is a territory technology with no practical ceiling and pricier optics; the cable itself is cheaper in single-mode, the labour is identical, and the plant outlives every optic you will ever plug into it. Links under a few hundred metres inside one building may honestly choose either — everything else is single-mode's by default. That asymmetry is not a sales position; it is what the reach table in every honest datasheet (550m against 40–120 km) has been saying all along. Choose the fiber for the life of the building, fit the module for the link in front of you, and write both choices on the drawing.
One last practical note on testing: whichever fiber you choose, the link is not finished when it lights. An OTDR trace or at minimum a power-meter reading at both wavelengths in use, recorded against the link's identifier, is the difference between a plant and a rumor. Re-measure after any building work near the route — the classic single-mode failure is not age but a backhoe, a crushed patch cord behind a rack, or a connector polished by dust and optimism. File the readings with the fiber record you were told to keep three sections ago; future you, standing in a riser at midnight with a meter in hand, is the audience for every word of it.