EDS is the easiest analytical technique in electron microscopy to run — and the easiest to run wrong. Park the beam, wait a few seconds, and the software hands you a spectrum, a peak list, and weight percentages with two confident decimal places. The detector always returns numbers. Whether those numbers mean what you think they mean is another matter entirely.
After enough years watching labs work, the same handful of mistakes show up over and over — usually invisible in the report, which is what makes them dangerous. Here are the seven we see most, and how to catch each one.
1. Running the wrong accelerating voltage for the elements in play
An X-ray line only appears if the beam energy is comfortably above the energy needed to excite it — as a working rule, aim for about twice the line energy. Iron’s Kα line sits at 6.4 keV, so a 10 kV beam will barely tickle it; you’ll be reading iron from its low-energy L lines, crowded down among the light elements, and your quantification will suffer for it. Steel work wants 15–20 kV.
The mistake runs the other direction too. At 30 kV on a polymer or a thin coating, the beam punches deep below the surface, so the spectrum describes bulk you may not care about — and light-element peaks ride on a worse background. Pick the voltage for the chemistry question you’re asking, not for the prettiest image. On an instrument like the SNE-Alpha, stepping from 1 to 30 kV takes seconds; there’s no excuse to run everything at one setting.
2. Trusting auto-ID without a fight
Peak identification software is a suggestion engine, not an oracle. EDS peaks are wide, and the periodic table is full of coincidences: sulfur Kα, molybdenum Lα, and lead Mα all crowd within about 50 eV of each other near 2.3 keV. Titanium Kβ lands on vanadium Kα. Manganese Kβ sits on iron Kα.
Auto-ID doesn’t know whether your sample came from a lubricant test, a solder joint, or a paint chip. You do. Every automated peak list deserves a human pass that asks: does this element make sense here, and could this peak be its neighbor instead?
3. Forgetting what you coated the sample with
A gold sputter coat is wonderful for imaging and a nuisance for EDS: gold’s M lines land near 2.1 keV, right in the neighborhood where phosphorus and sulfur live, and a coating a few nanometers thick still contributes real counts. We’ve watched “trace sulfur” findings evaporate the moment someone remembered the coater.
If elemental analysis is the goal, coat with carbon (which adds only a carbon peak — a problem only when carbon is your analyte), or skip coating entirely and run the sample uncoated in low-vacuum mode. Either way, tell whoever reads the report what was on the surface.
4. Quantifying rough surfaces as if they were polished
The correction models behind those tidy weight percentages — ZAF, φ(ρz) — assume a flat, polished specimen at a known take-off angle. A fracture surface violates every one of those assumptions: local topography shadows the detector, X-rays exit through wildly varying path lengths, and the reported composition can swing by several weight percent depending on which hillside you happened to click.
Qualitative identification on rough surfaces is fine. But when the actual numbers matter — specification disputes, failure analysis reports, anything with a customer attached — mount and polish a cross-section, or clearly label the result as semi-quantitative.
5. Assuming the beam only samples the pixel you clicked
The electron probe may be nanometers wide, but X-rays are generated throughout the interaction volume — which, at 20 kV in a light matrix, extends micrometers deep and wide. Point at a sub-micron particle sitting on a substrate and most of your spectrum is substrate.
Dropping the accelerating voltage shrinks the volume and pulls the analysis toward the surface — this is the same trade-off as mistake #1, worked from the other side. And if you’re running uncoated samples in low vacuum, remember the beam skirt: gas scattering means a slice of your signal comes from a wider neighborhood around the point of interest. It’s manageable — but only if you know it’s happening.
6. Letting the beam cook the sample
Some samples quietly change composition while you measure them. Sodium and potassium migrate away from the beam in glasses; polymers lose mass; hydrated or volatile phases evolve under dose. The spectrum you collect after thirty seconds of dwelling on one spot may describe a sample that no longer exists.
The defenses are simple: lower beam current, shorter acquisitions, and spreading the dose by scanning a larger area rather than parking on a point. If two consecutive acquisitions of the same region disagree, the beam is telling you something.
7. Overreading trace peaks — and underfeeding your statistics
Not every small peak is an element. EDS spectra contain sum peaks (two photons arriving together), silicon escape peaks (the detector’s own artifact, 1.74 keV below a parent peak), and system peaks — stray iron, copper, or aluminum excited from the stage, holder, or chamber rather than the sample.
And below all of that sits counting statistics. Under good conditions EDS detection limits are on the order of 0.1 weight percent; a flicker at half that level after a five-second acquisition is noise wearing a costume. Collect enough live time for the question you’re asking, and when a trace finding would change a decision, verify it: longer acquisition, a second location, or a blank stub as a sanity check.
The common thread
None of these mistakes come from bad detectors — modern silicon drift detectors like the Bruker XFlash 630 we pair with the SNE-Alpha are superb. They come from treating EDS as a button instead of a measurement. A few physics-aware habits — right kV, skeptical peak review, honest surfaces, adequate counts — are the difference between elemental analysis and elemental decoration.
Frequently asked questions
What accelerating voltage should I use for EDS?
Roughly twice the energy of the highest line you need to excite. For general metals work, 15–20 kV covers the useful K lines; for surface-sensitive or light-element questions, 5–10 kV keeps the analysis shallow — just expect to lean on L and M lines for heavier elements.
Why does my spectrum show gold (or palladium)?
Almost certainly your sputter coating. Coat with carbon when EDS matters, or image uncoated in low-vacuum mode and skip the question entirely.
Can EDS detect light elements like carbon and oxygen?
Modern windowless-capable and thin-window detectors see them readily, and mapping them qualitatively is routine. Quantifying them is harder — absorption effects and surface contamination both hit low-energy X-rays hardest — so treat light-element percentages with extra skepticism.
Want a second opinion on a spectrum?
If an EDS result doesn’t smell right — or you want to see what a properly configured benchtop SEM with EDS does on your actual material — send us the sample. We’ll run it live, show you every setting, and let the physics speak. Get in touch.