Every hard drive keeps a diary of its own decline — SMART data, logged constantly by the drive itself. The tools to read it are free and take two minutes. What most guides skip is the part that matters: which of the dozens of attributes actually predict failure, and what to do the day one of them moves.
Reallocated sectors. Pending sectors. Uncorrectable sectors. Spin retries. And the trend — whether any of them moved since last month. Everything else is context.
On Windows, download CrystalDiskInfo — free, tiny, and the de facto standard. It opens straight onto every drive’s SMART table with a coloured verdict up top: blue for good, yellow for caution, red for bad. On a Mac, Disk Utility shows a one-word SMART status under the drive’s info, and the smartctl command (via Homebrew’s smartmontools) gives the full table for those comfortable in Terminal. On any system, the drive maker’s own dashboard — Seagate SeaTools, Western Digital’s Dashboard — reads the same underlying data with a friendlier face.
One honest caveat before you trust any of them on an external drive: some USB enclosures don’t pass SMART data through, so a blank or ‘unsupported’ result on an external doesn’t mean healthy — it means unreadable through that cable.
Find these rows in the attribute table. Their raw values should be zero — and stay zero.
Reallocated Sectors (05) — patches of the disk that failed and were swapped for spares. The single most-watched number in the trade: a handful is survivable, a climbing count is a drive announcing its retirement. Current Pending Sectors (C5) — sectors the drive can no longer read reliably and is waiting to judge; these are often your files, currently unreadable. Any non-zero value here means back up now. Uncorrectable Sectors (C6) — reads that failed even after every retry: confirmed damage. Spin Retry Count (0A) — the motor needed more than one attempt to spin up; a mechanical red flag far rarer and far more serious than it looks. And fifth, the trend: screenshot the table today, compare in a month. A drive with three reallocated sectors forever is tired; a drive that went from zero to three to eleven is failing on a schedule.
Power-on hours and temperature give context — a five-year, 40,000-hour drive deserves more scepticism — but they predict nothing by themselves.
The instinct when a drive shows caution is to run more scans — a full surface test, CHKDSK, a vendor ‘repair’. Resist it. Every deep scan is hours of sustained reading on hardware that just told you it’s struggling, and pending sectors have a habit of multiplying under exactly that load. The right order: copy your irreplaceable files off first, most important folders leading, while the drive still reads; then decide the drive’s fate at leisure. If the copying itself stalls, hangs the machine, or the drive starts making any new noise — clicking, buzzing, repeated spin-downs — stop entirely and power off: that’s the point where every further minute of running costs data, and where clicking-drive recovery on a bench, imaging the drive on its own terms, is the safe road. A drive that already failed — not detected, wrong size, endless hangs — skips the checklist and goes straight to hard drive recovery.
External drives complicate the check in one specific way: the USB bridge inside many enclosures doesn’t pass SMART commands through, so your tool shows a drive with no health data at all. Don’t read that as a clean bill — it’s a language barrier, not a verdict. Three workarounds, in rising effort: try a tool with better bridge support (CrystalDiskInfo reads more enclosures than most), try the drive in a different dock or caddy, or — for a drive that matters and misbehaves — let the diagnostic bench read it natively. And a listening note specific to externals: their small enclosures amplify sounds, so a healthy external often sounds ‘louder’ than an internal drive; it’s rhythm and character that signal trouble, not volume alone.
No — and this surprises people. CHKDSK checks the file system’s logic, not the drive’s physical health, and it does so by writing corrections. On a mechanically weakening drive, its long, punishing read-and-repair passes are the opposite of a check-up: they’re a stress test with a pen. Read the SMART data instead — it’s the drive’s own health record, gathered passively.
Treat the percentage as marketing and read the numbers behind it. A drive can show ‘95% healthy’ while its reallocated-sector count is climbing week on week — the single strongest predictor of failure. The values that matter reading zero matters far more than any headline score; the five attributes above are the ones an engineer actually looks at.
The clicking, every time. SMART only records what the drive’s own electronics can measure and report, and a mechanical fault in the heads can sit outside that entirely — right up until the failure. Any clicking, buzzing, or repeated spin-up-and-stop overrules any passing score: power the drive down and treat it as failing.
Context, not verdict. Drives are comfortable well into the 40s Celsius, and a warm summer afternoon under load isn’t a symptom. What matters is change and extremity: a drive that suddenly runs 15 degrees hotter than its own history, or one living above 55°C, is either working far harder than it should (retry storms have a heat signature) or being cooked by a failed fan — both worth investigating. Log the temperature with each health check so ‘hotter than usual’ is a measurement rather than a feeling.
Free diagnostic at our Belfast lab: an honest read of what your drive is telling you, and a written quote before any recovery work if it comes to that.