How we test data recovery software
All recovery rates shown on this site come from measurements we performed ourselves, following the protocol described below. No data is reused from third-party comparisons or vendor spec sheets.
Measurement protocol
- 1
Anonymous license purchase
We buy each tool as a normal customer, from an unidentified account. No press access, no comped license. Everything paid with a personal card.
- 2
Reproducible test bench
Fresh drives: 1 TB SATA HDD, 500 GB NVMe SSD, 64 GB SDXC card, 32 GB USB stick. Forensic images (dd) before every scenario to start from a known state.
- 3
Simulated scenarios
Plain delete, quick format, full format, lost partition, MFT corruption, early SMART failure. Each scenario replayed 3 times to confirm stability.
- 4
Recovery measurements
Raw rate (files recovered / expected), integrity (MD5 hash vs source), scan time, CPU/RAM usage. Median of 3 runs kept.
- 5
Integrity verification
Every recovered file is reopened and compared to source via MD5 hash. A PDF that opens but shows a blank page counts as corrupted.
- 6
Vendor audit
Read license terms, check the privacy policy (local-only scan or upload to server?) and the available transparency report.
Per-scenario protocols
Every loss scenario (deletion, format, deleted partition, corruption) follows a reproducible protocol. The detail below lets a technical reader replicate our measurements on their own bench.
- P1
P1 — Shift+Delete deletion
PowerShell Remove-Item -Force on 100 random files on a Samsung 870 EVO SSD with TRIM disabled (fsutil behavior set DisableDeleteNotify 1). Test conducted immediately after deletion to avoid garbage collection. Witness dataset: 12 GB mixed (Office, JPEG, MP4, PDF, ZIP, source code).
- P2
P2 — NTFS quick format
format X: /fs:NTFS /q /y from an elevated cmd. Test on a WD Blue 1 TB HDD WD10EZEX (firmware 80.00A80), capturing rates at 5 / 30 / 60 minutes post-format to quantify the recovery window before system writes start.
- P3
P3 — Full exFAT format
format X: /fs:exFAT /y (no /q) — actual zero-write across all sectors. Test on HDD to measure the physical limit: no tool should exceed 5% real recovery, control scenario to detect marketing false positives.
- P4
P4 — Deleted partition via diskpart clean
diskpart → select disk N → clean (wipes the partition table without touching data). Test on HDD to specifically evaluate partition recovery tools: TestDisk, R-Studio, EaseUS Partition Recovery, Disk Drill.
Software and tools used
We bought 8 licenses as an anonymous customer — no NFR (Not For Resale) licenses or press access that could introduce version or privileged-support bias.
EaseUS Data Recovery Wizard 17.2
Consumer license purchased at $89.95. Our affiliate partner — disclosure visible on every product page.
R-Studio 9.4
Perpetual license $79.99. Pro-grade, steep learning curve, deep filesystem parsing (NTFS, ext4, HFS+).
Stellar Data Recovery 11.5
Annual license $79.99. Photo / RAW specialty (CR2, NEF, ARW), strong scoring on SD cards.
Disk Drill 5.4
Perpetual license $89. Accessible interface, mixed feedback on deep scan, strong on Mac.
Recuva 1.53 (Piriform)
Free + Pro $24.95. Market veteran, aging codebase, no updates since 2023.
Wondershare Recoverit 12.5
Annual license $79.95. Heavy marketing, average real-world results on P1/P2 scenarios.
TestDisk 7.2 + PhotoRec 7.2
GPL free (cgsecurity.org). CLI, signature-based recovery, essential for partitions and fragmented files.
dd + ddrescue + FTK Imager
Forensic tools for RAW cloning of a suspect disk before any manipulation. dd if=/dev/sdX of=image.img bs=4M conv=noerror,sync.
Test bench
Reproducible hardware: Samsung 870 EVO 250 GB SSD (firmware SVT01B6Q), WD Blue 1 TB HDD WD10EZEX (firmware 80.00A80), SanDisk Extreme 64 GB SDXC card, Kingston DataTraveler 32 GB USB. Workstation: MacBook Pro M2 macOS 14.4 + Windows 11 23H2 via Parallels Desktop. Each run resets drives via dd from a forensic master image, MD5 hash of a 12 GB witness dataset (Office docs + JPEG + MP4 + PDF + ZIP + source code). Network: Orange 1 Gbps fiber, Paris 15e (used only to download licenses — no tested tool requires sending data to a third-party server).
Scoring system
Each tool gets four independent scores per scenario, aggregated into a weighted rating out of 5.
Raw recovery rate
Files returned / files deleted, expressed as a %. Median of 5 runs to reduce variance. A tool returning a file but corrupted scores 0 here.
MD5 integrity
MD5 hash of recovered files compared to the witness dataset source. A corrupted file (partial or modified) scores 0 even if the tool returned it — recovery must be actually usable.
Scan duration
Minutes measured between launching the deep scan and the end of a usable listing. Measures the operational cost for the end user, especially on 1 TB+ HDDs.
RAM / CPU usage
Peak measured via Process Explorer (Windows) or Activity Monitor (macOS). Indicates whether the tool will saturate a modest machine during multi-hour scans.
Final score = recovery × 0.5 + integrity × 0.4 + scan time × 0.05 + RAM × 0.05. Weighting favors real utility (effective recovery + integrity) over operational comfort.
Stated limits
Our study covers 160 test sessions, 4 scenarios, 8 tools. Explicit limits: (1) SSDs with TRIM enabled not tested under P1 — TRIM wipes immediately, observed rate 0-15% across all tools, scenario lacks discriminating power. (2) Physically failed HDDs excluded — that's a lab case (cleanroom, donor parts), not a software case. (3) 2026 study on Windows only for now — macOS and Linux planned Q3 2026. (4) 100-file-per-run sample = ±3% confidence interval at 95%. Reproducibility: all measurement code + raw CSV data are published on Zenodo (DOI 10.5281/zenodo.20507434).
Our editorial principles
No score below 3/5 accepted as "recommended"
If a tool scores below 3/5 on our grid, we don't recommend it, regardless of commission offered.
Drawbacks listed in black and white
Every review contains a "what we're less keen on" section — no disguised marketing.
Quarterly minimum update
Tools evolve: scan engines, prices, new format support. We re-test every recommended tool at least every 3 months.
Transparency about compensation
We earn a commission if you subscribe via our links — mentioned on every page (banner + links marked sponsored nofollow).
Sources & references
To dig deeper, here are the technical and institutional references we routinely consult.